<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/">
  <channel>
    <title>Technical Chops by Simon Minton</title>
    <link>https://www.technicalchops.com</link>
    <description>Insightful analysis and in-depth, easily digestible understanding of the medium to long-term effects of emerging technologies.</description>
    <language>en</language>
    <lastBuildDate>Tue, 22 Sep 2026 15:30:48 GMT</lastBuildDate>
    <atom:link href="https://www.technicalchops.com/rss.xml" rel="self" type="application/rss+xml"/>
    <item>
      <title>Taking Their Word For It</title>
      <link>https://www.technicalchops.com/articles/taking-their-word-for-it</link>
      <guid isPermaLink="true">https://www.technicalchops.com/articles/taking-their-word-for-it</guid>
      <description>AI, Covid, and what happens when we start checking the experts</description>
      <content:encoded><![CDATA[<p>For as long as any of us have been alive, you trusted the expert. You went to the doctor, or the solicitor, and they told you what was what, and you took their word for it. You didn’t really have a choice. You didn’t know enough to argue, you’d no way of checking, and questioning the professional wasn’t the done thing anyway. So you nodded, and got on with it. That deference held a surprising amount together: a kind of trust we’ve handed to institutions for decades without thinking about it much.</p>
<p>That’s starting to come apart, and not only because of AI. Faith in the experts has been wobbling for years - Covid did a real number on it, with the whole country watching them contradict each other and change their advice in real time. Although that cut both ways. Plenty of the people who decided to ignore the experts altogether didn’t exactly cover themselves in glory. Sometimes the expert really is right, and the confident bloke online really isn’t.</p>
<p>AI does something different, though. It lets you actually check. You come out of the appointment, go home, and ask it what it reckons about what you were just told. Loads of people are already doing exactly this. It feels like a small thing, but it isn’t: you’re second-guessing the professional in a way that simply wasn’t available a few years ago, for free, on your phone, with nobody to feel awkward in front of. And some of the time it’s genuinely useful. Experts get things wrong, the symptom waved away or the advice that was a bit off, and now you can at least ask the question you’d never have known to ask.</p>
<p>It isn’t all good, though. There’s something a little corrosive about a world where nobody quite trusts the professional any more, where the doctor who’s right ninety-nine times gets fact-checked on the hundredth. And the thing we’re checking with isn’t exactly trustworthy itself; it’ll give you something completely wrong in the same confident voice it uses for everything.</p>
<p>The AI itself is built partly out of doctors, and not average ones. The companies behind these things pay top people in each field, leading doctors and lawyers and the rest, to sit and write the answers that train the models. So when you check your GP against it, you’re in a way reaching past them to the best in the field. That’s not nothing. But you get it all as one flat, confident summary, with the bit that tells you how much to trust any given part of it taken out.</p>
<p>I don’t think there’s a clean answer here. It’s a good thing and a bad thing at once. We spent a very long time trusting the people in the room because we had no real alternative. Now we have one, of sorts. I just can’t tell yet whether it leaves us better off, or simply more alone with decisions we were never that well equipped to make in the first place.</p>
<p><em>Originally published at <a href="https://www.technicalchops.com/articles/taking-their-word-for-it">technicalchops.com</a>.</em></p>]]></content:encoded>
      <pubDate>Tue, 30 Jun 2026 00:00:00 GMT</pubDate>
      <author>simon@minton.co (Simon Minton)</author>
      <category>AI</category>
      <category>Trust</category>
      <category>Institutions</category>
      <category>Expertise</category>
    </item>
    <item>
      <title>Buying the Badge</title>
      <link>https://www.technicalchops.com/articles/buying-the-badge</link>
      <guid isPermaLink="true">https://www.technicalchops.com/articles/buying-the-badge</guid>
      <description>Ferrari&apos;s Luce, EVs and the one thing a Tesla can&apos;t sell</description>
      <content:encoded><![CDATA[<p>Good morning,</p>
<p>A contrarian take to go with <a href="https://www.technicalchops.com/articles/chasing-cool-jaguars-gamble">the Jag’s</a> — this time on Ferrari’s much-maligned electric Luce.</p>


<p>Well. I commented on the Jag with a contrarian take, so it feels only fitting that I do the same for Ferrari’s effort.</p>
<p>The Luce’s reception has been <em>brutal</em>. The share price fell close to 8%, Luca di Montezemolo, who ran the company for twenty-three years, said they should take the prancing horse off it, and even Matteo Salvini, the Deputy Prime Minister and a reliable booster of Italian industry, would not defend it. The design has been compared to a Honda Accord, a Nissan Leaf, and a Tesla Model 3 that someone sat on. The general view is that Enzo would be turning in his grave.</p>
<p>He probably would be. But the styling is probably the least interesting thing about this car. What the Luce is designed to do is break a rule that has held fast for a decade, and the clearest way to see it is to look at the company going the other way.</p>

<p>Ben Thompson made the case this week that Tesla proves something true of all modern technology: the rich and the rest end up with the same thing. The richest people he knows, he wrote, drive Teslas not for the finishes, which are cheap, but for the self-driving software, which nothing else can match. But the car a billionaire buys to drive for them is the same one half the mums in London use for school drop off.</p>
<p>The principle runs through everything with a chip in it. Tim Cook does not own a secret gold iPhone; he owns the one you own. There is no executive edition of the AirPods, no billionaire’s iPad. Whatever the best phone or pair of headphones happens to be, the wealthiest person alive owns the same one you can. Technology is the one luxury money cannot upgrade.</p>
<p>Cars were never that. A car is the most conspicuous thing most people will ever own, and the gap between the cheapest and the most expensive has always been vast and deliberate. But an electric car is barely a car in the old sense. Strip the body and it is a battery, some motors and a great deal of software, and software levels everyone. These aren’t cars; they are computers with wheels.</p>
<p>So for a decade the electric car turned as egalitarian as the phone. There were grander electric cars – a Rolls-Royce Spectre, a Maybach EQS – but the Rolls is built to waft and the Maybach to be chauffeured; neither is the car you buy as an ultra-premium daily driver. The best electric car you could drive hard was a Tesla Plaid, and a Tesla Plaid is what a Silicon Valley software engineer buys. If you were seriously rich and wanted to drive something electric, your shortlist was the same as someone earning a tenth of your salary: the Tesla, a Taycan, a Lucid or an electric BMW.</p>
<p>This is a real problem for a particular kind of buyer – the kind who cares a great deal about the statement his car makes. He might well even have a V12 Ferrari in the garage for the weekend, but he wants something electric for the week, and right now the electric car in his driveway is the same one his head of engineering parks next to his at work. He does not want that. He has never wanted that. For everyone else the levelling of the car is a curiosity; for him it is a daily irritation he has the money to fix and, until now, no way to.</p>
<p>The Luce is Ferrari’s fix: the first fast electric car that also makes a statement. For the first time, money buys you a better electric car to drive – and that, not the styling, is the entire point of it.</p>
<p>And – this is the part the styling row has drowned out – it is apparently a genuinely extraordinary car to drive. There is a motor on each wheel and around 1,050 horsepower between them; 62 mph arrives in 2.5 seconds, 200 km/h in under seven. But the raw figures are the least of it. With no conventional gearbox, Ferrari hands the driver two paddles that do something no combustion car ever could: pull the right and the four motors release their full torque for a corner exit, pull the left and they trade it for regenerative braking – the control unit re-aiming all of it two hundred times a second to keep well over two tonnes flat and gripping. It is torque vectoring of a precision a mechanical drivetrain cannot come close to.</p>
<p>And where almost every other maker now pipes a synthetic engine note through the speakers to paper over the silence, Ferrari does the opposite: it amplifies the Luce’s own sound, the real frequencies of its motors and electronics, and leaves it there. It is a small thing that says a large one. This is not a company faking the old Ferrari; it is a company deciding what the next one sounds like.</p>
<p>The objection to the Luce comes in two parts: that it is ugly, and that it is not “a Ferrari”. It is both. But the two have different causes, and only one of them is Ferrari’s fault. Start with the part that isn’t. A combustion Ferrari is dramatic because it is allowed to be; an electric one is not. This is a piece of technology, and the technology dictates the shape. At around 2,300 kilos the Luce is the heaviest Ferrari ever made, heavier even than the Purosangue, and most of that weight is a slab of cells laid flat under the floor. That mass fixes the proportions before a designer draws a line: the long wheelbase, the high body, the short overhangs. Then there is the air. Range is the only number that matters in an electric car, and at speed the enemy of range is drag. The Luce has the lowest drag coefficient of any Ferrari ever made, 0.254, against roughly 0.34 for the F40. Everything that makes a combustion Ferrari beautiful – the low nose, the deep intakes, the splitters, the wing – exists to manage air at speed, and every one of them creates drag. None of it survives the switch to electric, and that was set before Jony and Marc first put pencil to paper.</p>
<p>Ferrari’s chief designer, Flavio Manzoni, said it plainly: on a combustion car you design for downforce, on an electric one you design for drag, and the body “must be almost continuous, almost flat. Even the wheels must be flat.” That is the constraint.</p>
<p>It is why fast electric cars have all drifted toward the same general silhouette – the Tesla, the Lucid, the Taycan, the Mercedes EQS, each with the same big, long fastback, the same flush wheels, the same smooth, intakeless face. They did not copy one another. They were designed by the same maths. Until battery chemistry improves<sup id="fnref-1"><a href="#fn-1">1</a></sup> enough that range no longer dictates the shape, every fast electric car will look broadly like this one.</p>
<p>This is also the answer to the obvious objection, which is that plenty of electric cars look good. Some do. But the most dramatic ones – the Chinese electric supercars, say – manage it by not caring about range, a freedom a car meant to be lived with does not have. The handsome ones that are meant to be lived with – basically only the Taycan – are handsome within the same smooth envelope, not in defiance of it. Drama and range pull in opposite directions, and Ferrari built the Luce to be driven on the road, not the racetrack.</p>
<p>The Luce cannot win on looks, then, and it does not need to. It is not competing with the F80, the 12Cilindri, or any car with cylinders – Ferrari has no plans to stop building those. It is competing with the Tesla and the Taycan, where it holds an advantage no rival can engineer. A Porsche is a driver’s car; a Ferrari is a <strong>statement</strong>, and at this level the statement is what the money buys.</p>
<p>And then the same logic, multiplied, points east. The buyer I have described exists in London and California, but there are far more like him in Shanghai, where Ferrari has never properly cracked the market and no one feels much nostalgia for a V12 they were never really sold. Nor is there much patience there for quiet luxury: the Luce wears no fewer than eight prancing horses, with a light-up key as a ninth – the kind of thing that “accidentally” falls from a pocket onto a restaurant table. In the land of the logo, that is a feature, not an embarrassment. The tax seals the deal. China punishes big engines – a displacement levy reaching 40%, before import duty and VAT – potentially doubling the cost of the car – so a combustion Ferrari costs far more there than at home, while an electric one skips the engine tax entirely.</p>
<p>The other charge – that it is simply ugly – is the fair one. Ive and Newson’s rounded minimalism suits a phone, a thing you are meant to forget you are holding; but a $500,000 car is meant to be looked at. The colours are loud, but the surfaces beneath them are forgettable, and forgettable is the one thing an object bought to be seen cannot be. The envelope was forced on Ferrari; the dullness was not. They had room to do more within it, and didn’t.</p>
<p>Tesla killed off the Model S and X this year and handed the factory to its robots; its richest customers now have the same cars as everyone else. Ferrari has gone the other way, and built the rich a car the rest of us cannot have. The Luce is not really selling a car. It is selling the one thing a Tesla cannot: the badge.</p>
<hr class="mt-8 border-gray-200">
<p class="text-sm text-gray-500" id="fn-1">1. The chemistry that will eventually change that – solid-state cells, silicon anodes – is a good part of what we focus on over at <a href="https://canaryiq.com">CanaryIQ</a>. <a href="#fnref-1">↩</a></p>

<p><em>Originally published at <a href="https://www.technicalchops.com/articles/buying-the-badge">technicalchops.com</a>.</em></p>]]></content:encoded>
      <pubDate>Fri, 29 May 2026 00:00:00 GMT</pubDate>
      <author>simon@minton.co (Simon Minton)</author>
      <category>Auto</category>
      <category>Cars</category>
      <category>EVs</category>
      <category>China</category>
    </item>
    <item>
      <title>Transparent Aluminium</title>
      <link>https://www.technicalchops.com/articles/transparent-aluminium</link>
      <guid isPermaLink="true">https://www.technicalchops.com/articles/transparent-aluminium</guid>
      <description>AI is mapping every material we could have. Making them is another matter.</description>
      <content:encoded><![CDATA[<p>There’s a scene in Star Trek IV - the one where they go back to the eighties to fetch a couple of whales - where Scotty needs a huge tank built and hasn’t got the materials for it. So he sits down at a clunky old Mac, tries talking to it, gives up, picks up the mouse like a microphone, and in the end just types out the formula for “transparent aluminium” in exchange for the plexiglass he’s after. The manufacturer stares at the screen, baffled, while Scotty cheerfully hands over a material that won’t be invented for a few hundred years.</p>
<p>That scene stuck with me as a kid, more than the phasers or the warp drive ever did. Something about the idea that the really futuristic thing wasn’t a machine at all, but a material - a substance we just didn’t have the recipe for yet.</p>
<p>I keep thinking about it lately, because we’re getting close to the first half of that. Not the chatbots, but a different kind of AI - the sort that models physics and chemistry rather than language. It’s started working through the space of materials that could exist: which combinations of elements would actually hold together, which crystals are stable, which steels and glasses are perfectly possible but nobody has ever happened to mix. The numbers are staggering. People are talking about hundreds of thousands of stable materials we’d never catalogued, found not in a lab but on a computer, by working out what nature will and won’t allow.</p>
<p>So the recipe part is, slowly, arriving. We’re going to know what’s possible.</p>
<p>The catch is that Scotty didn’t just name the stuff. He had to hand over how to make it, and even with the formula in front of them, that eighties manufacturer would have spent years and a fortune actually producing it. That’s the part the film skips over, and it turns out to be the part that matters. Knowing a material is possible is nothing like being able to make one. Predicting that something is stable is the easy bit next to working out the temperatures, the pressures, the order of the steps, the fiddly craft of getting atoms to line up like that out in the real world.</p>
<p>And, as it happens, transparent aluminium is real. There’s a ceramic called ALON that gets sold, fairly straight-facedly, as “transparent aluminium armour”. The reason you’re not looking out of an aluminium window right now isn’t that nobody knows it’s possible. It’s that making the stuff is slow, hard and expensive.</p>
<p>Which is roughly where I think we’re going: a growing list of miracle materials we know are possible and mostly can’t make yet. A library of the future, running miles ahead of the factory. The hard part stops being what could exist and becomes the duller question of how on earth (or <a href="https://www.nasa.gov/missions/station/iss-research/optical-fiber-production/">perhaps not</a>) you actually build it. Scotty got the formula across in about a minute. The rest of it was always going to be the real work.</p>
<p><em>Originally published at <a href="https://www.technicalchops.com/articles/transparent-aluminium">technicalchops.com</a>.</em></p>]]></content:encoded>
      <pubDate>Wed, 08 Apr 2026 00:00:00 GMT</pubDate>
      <author>simon@minton.co (Simon Minton)</author>
      <category>AI</category>
      <category>Materials</category>
      <category>Science</category>
      <category>Manufacturing</category>
    </item>
    <item>
      <title>26x</title>
      <link>https://www.technicalchops.com/articles/26x</link>
      <guid isPermaLink="true">https://www.technicalchops.com/articles/26x</guid>
      <description>Mind. Blown.</description>
      <content:encoded><![CDATA[<p>I’ve started telling people that I am working at 26x speed. It sounds like hyperbole, which is sort of deliberate. But it’s not. Since the beginning of the year, I have done six months of work – of coding – every week.</p>
<p>Working with AI agents, and especially Claude Code, has changed my entire coding life.</p>
<p>I’ve always been driven to build, and describe myself as a builder on my <a href="https://www.minton.co">personal page</a>, but I’ve often become frustrated with how long it takes to do the things required to build something great. Even a simple CRUD app takes a decent amount of footwork. Add an item. Delete an item. Edit an item. Add validation. Build relationships between items. That’s hours of work.</p>
<p>I would get bored implementing all of the basic states. It would run my battery down, even as I pushed through. The incremental dopamine hits of a function working as I hoped were sometimes enough – but there came a point where I just couldn’t do any more for the day. And I consider myself to be pretty persistent.</p>
<h2 id="5x10x"><strong>5x–10x</strong></h2>
<p>Then came the first “good” versions of LLM code writing. Cursor. It would autocomplete, correctly, what I wanted to do more often than not. I started chatting and asking it to build out features, iterating back and forth until we got to where we wanted to be.</p>
<p>By the middle of last year I had become the coder I wanted to be. The tooling had made it possible for me to complete around a week of 2024 work in a day – albeit with the caveat that most of my time was spent checking the work, with occasional bouts of huge frustration as the LLM did something completely inane.</p>
<p>Codex was a game changer. Moving to Codex from Cursor meant that I could finally level up to being a 10x coder. Two weeks of work in a day. We need a micro app to solve a very specific internal problem? No worries, just give me a day.</p>
<p>My output was high, but I still sank a lot of my time into making sure that the code was right and really tweaking designs. It was great, but it wasn’t <em>great</em>.</p>
<h2 id="and-then-came-claude"><strong>And then came Claude</strong></h2>
<p>I had used Claude Code a little previously but had stuck to the OpenAI ecosystem – the tools were pretty comparable. Until an update in December, when everything changed. I saw murmurings on X and decided to play a little over the Christmas break. It was an entire reality shift.</p>
<p>Spec out what you want built, and it would build it. End to end. No laziness. Correct integrations. It would actually go and get the documentation I asked it to. It removed all of the <strong>effort</strong> involved in building an app.</p>
<p>Ideas that have sat on the shelf for years are built and live. In the course of a month I have done more than I could have done in well over a year. It has blown my mind. My problems are no longer code related. They are all <em>actual business</em> problems.</p>
<p>This year I will produce more code – more features – than I could have been expected to in my entire career just five years ago. Which is just… wild.</p>

<p><em>Originally published at <a href="https://www.technicalchops.com/articles/26x">technicalchops.com</a>.</em></p>]]></content:encoded>
      <pubDate>Sun, 01 Feb 2026 00:00:00 GMT</pubDate>
      <author>simon@minton.co (Simon Minton)</author>
      <category>AI</category>
      <category>Musings</category>
      <category>LLMs</category>
    </item>
    <item>
      <title>The Models Don&apos;t Matter Anymore</title>
      <link>https://www.technicalchops.com/articles/the-models-dont-matter-anymore</link>
      <guid isPermaLink="true">https://www.technicalchops.com/articles/the-models-dont-matter-anymore</guid>
      <description>Frontier releases keep coming, but for businesses, focusing on flexible infrastructure pays bigger dividends than chasing the latest and greatest.</description>
      <content:encoded><![CDATA[<p>Another day, another frontier-level release—welcome to the world, <em>Claude 3.7 Sonnet</em>. And yet, the churn of model announcements increasingly feels like background noise. For the past two years, conversations about AI have fixated on which model is best: who has the most parameters, who trained on the biggest cluster, whose benchmarks are highest… But do these improvements really matter anymore?</p>
<p>Ordinarily, when an article poses a rhetorical question, the safe bet is: “<em>No</em>.” But that’s not actually the case here—new models <em>do</em> matter. Yet for most businesses and consumers, the impact today is modest at best.</p>

<p>The constant stream of model releases are the result of the “great race”. OpenAI, Google, Anthropic, xAI, Mistral, DeepSeek, et al. are all chasing what they hope will be <em>the</em> model—the one capable of AGI (Artificial General Intelligence). Definitions vary. <a href="https://openai.com/charter/">OpenAI’s charter</a> describes AGI as systems that “outperform humans at most economically valuable work.” Sam Altman recently <a href="https://blog.samaltman.com/">tweaked this definition</a> to something like “a system that can tackle complex problems at human level, across many fields.”</p>
<p>My own definition is more specific:</p>
<p><strong>The first true AGI is the first model that can, with limited human input, build a better version of itself.</strong></p>
<p>In fundamental research then, the drive for better models is far from meaningless. Performance gains, more elegant architectures, and greater efficiency keep nudging us closer to Altman’s definition of AGI—if not OpenAI’s or my own. Specialized models also matter, and matter a lot: Google’s “co-scientist” model has shown real promise (<a href="https://www.biorxiv.org/content/10.1101/2025.02.19.639094v1">paper</a>), replicating in days hypotheses that scientists themselves honed over a decade of research (<a href="https://research.google/blog/accelerating-scientific-breakthroughs-with-an-ai-co-scientist/">blog</a>).</p>
<p>But for most businesses—and for most of us—this doesn’t (and shouldn’t) carry that much weight.</p>
<hr>
<h2 id="dont-let-perfect-be-the-enemy-of-good">Don’t Let Perfect Be the Enemy of Good</h2>
<p>When LLMs first arrived, the experience was mind-bending. The GPT-3.5 generation could generate coherent prose so convincing that a <a href="https://www.scientificamerican.com/article/google-engineer-claims-ai-chatbot-is-sentient-why-that-matters/">Google engineer believed their system had achieved sentience</a>. It was natural to see those emergent abilities as a quantum leap in machine intelligence, because, well, they were.</p>
<p>The jump from GPT-3 to GPT-4, Claude 3.5 Sonnet, Google’s Gemini, and the like felt like another quantum leap. With the right coaxing, they became <em>really</em> good. Now, in the newest wave—OpenAI’s “o-series,” DeepSeek’s R1, and Grok 3—we again see models that are …<em>really good.</em> The leaps are getting harder to spot, not because they aren’t happening but because for <em>most tasks</em> and <em>most people,</em> existing LLMs already do the job.</p>
<p>For standard day-to-day tasks that require college-level writing or problem-solving, today’s models meet the bar. The difference between, say, <em>Claude 3.7 Sonnet</em> and what might be released next month is real, but often subtle. The game-changing improvements have already happened.</p>
<p>My argument is that the current crop of models are the ones businesses should build around—regardless of what’s coming next, and <em>even if they don’t always beat human performance right now.</em> By the time your project goes live, the model you spec’d on Day 1 will be replaced by a faster, cheaper, and more capable successor anyway.</p>
<hr>
<h2 id="preparing-the-groundwork">Preparing the Groundwork</h2>
<p>The key is to be model-agnostic. Treat the model like a new employee who will be “upskilled” every few months. Don’t design systems that rely on the quirks of a single LLM; assume the best option will shift frequently.</p>
<p>I’ve spent the past couple of months thinking about general rules for working with AI:</p>
<h3 id="1-build-to-be-model-agnostic">1. Build to Be Model-Agnostic</h3>
<p>It’s safe to assume that models will keep improving and that different models excel at different sub-tasks. Your system should let you swap between them easily, upgrade seamlessly, and test performance quickly. Create well-structured pipelines so you can deploy cheaper models for high-volume tasks, then tap a more powerful model only when complexity demands it. It’s the same principle we use at <a href="https://www.canaryiq.com/">CanaryIQ</a>: we allocate tasks to whichever model handles them best. This flexibility is crucial given how fast the AI landscape evolves and how new capabilities appear “overnight”.</p>
<h3 id="2-focus-on-your-data-integrations">2. Focus on Your Data Integrations</h3>
<p>Data remains the ultimate determinant of AI success. Even the most advanced model will falter if your data is inaccurate, incomplete, or poorly labeled. Conversely, an average LLM can shine with well-structured and reliable data. Many organizations discover that data engineering is the real bottleneck. Cleaning, labeling, and governing your data often yields bigger returns than anything else–and this can be done using current gen LLMs.</p>
<h3 id="3-fine-tuning-models-is-probably-not-worth-it">3. Fine-Tuning Models Is (Probably) Not Worth It</h3>
<p>Once upon a time, fine-tuning was the standard for customizing a model to a specific domain. But in an era where a new, frontier-level release appears every few months, that strategy can quickly become obsolete. Why spend loads of engineering effort and dollars on a fine-tuned model that might be overtaken by a more advanced general model tomorrow?</p>
<p>As context windows expand, you can simply feed more relevant data into the prompt itself, sidestepping the need for complex retraining. This approach makes it easier to pivot as soon as a better model arrives.</p>
<h3 id="4-get-good-at-prompt-engineering">4. Get Good at Prompt Engineering</h3>
<p>Effective prompt engineering can embed domain expertise without entrenching your workflows in a single model. In many cases, LLMs themselves can help generate optimized prompts for specific tasks. You can automatically iterate through variations, test performance, and integrate the best prompts based on internal benchmarks. By treating prompt engineering as an ongoing process, you keep your system nimble and resilient.</p>
<h3 id="5-orchestration-memory-and-retrieval-augmented-generation">5. Orchestration, Memory, and Retrieval-Augmented Generation</h3>
<p>Real-world tasks often go beyond a single question-and-answer step. You may need to retrieve documents, compile them into a prompt, verify outputs, or query secondary data sources. This multi-step chain requires orchestration frameworks that let you slot in different LLMs as needed. If your logic isn’t bound to the quirks of one model, it’s easy to upgrade whenever a better contender shows up.</p>
<h3 id="6-agentic-workflows-and-tool-use">6. Agentic Workflows and Tool Use</h3>
<p>Automated “AI agents” that chain tasks, call APIs, and pursue complex goals are the biggest news at the moment. But they’re only as useful as the systems behind them—reliable data access, robust orchestration, and well-defined processes. The choice of LLM matters far less if the rest of your infrastructure is brittle.</p>
<h3 id="7-models-as-commodities">7. Models as Commodities</h3>
<p>LLMs are commoditizing. The leading-edge model might shine for a particular challenge, but only until the next iteration arrives. You can’t simply chase “the best” model indefinitely. Instead, build flexible platforms that let you test new releases, confirm improvements, and switch as soon as it makes sense.</p>
<h3 id="8-disposable-software">8. Disposable Software</h3>
<p>The AI market moves at ludicrous speed. There’s little point in an 18-month build cycle if everything is outdated the day you ship. This doesn’t mean writing sloppy code, but it does mean building in a way that expects short life cycles. Your architecture should let you drop in a new LLM, run tests, confirm upgrades, and go live with minimal reengineering.</p>
<hr>
<h2 id="looking-ahead">Looking Ahead</h2>
<p>Models will continue to matter—and they’ll keep hitting the front-page of hacker news. At some point, we may even cross the threshold of AGI by my own definition (a system that can iteratively improve itself) and all hell may break loose. But for most organizations, whether that milestone arrives next year or the next decade is noise.</p>
<p>Focus on what you can control: a flexible infrastructure that pivots whenever a better model appears. That’s the essence of future-proofing your AI strategy. Don’t obsess over which LLM is hot this month—just be ready for whatever’s next. The models keep changing, but the right structures will outlast every frontier release.</p>
<p>*This article came from a talk I’ve given on this subject - the last version is <a href="https://www.technicalchops.com/documents/building-for-ai_technical_chops.pdf">here</a></p>


<p>Because the AI market moves faster than any single model’s development cycle. Building around a specific LLM risks missed opportunities as better options emerge. By embracing a model-agnostic strategy, you future-proof your AI efforts, reduce implementation headaches, and keep your focus on what truly drives results: data quality, flexible infrastructure, and the ability to pivot as soon as new capabilities arise. In short, the real competitive advantage doesn’t come from chasing the best model—it comes from designing systems that can absorb whatever comes next.</p>


<ul>
<li><strong>Model-Agnostic Infrastructure</strong> Build systems that can swap between LLMs seamlessly, ensuring quick upgrades and minimal disruption when new models arrive.</li>
<li><strong>Focus on Data Quality</strong> Clean, well-labeled, and accurately governed data can make even mid-range models shine, while poor data can sink any AI effort.</li>
<li><strong>Fine-Tuning Is Overhyped</strong> Constantly emerging frontier models often outpace costly fine-tuning.</li>
<li><strong>Prompt Engineering Matters</strong> Well-crafted prompts can embed domain knowledge and outperform heavily customized models—often with minimal overhead.</li>
<li><strong>Prepare for Rapid Change</strong> Software and strategies that take 18 months to deploy risk obsolescence. Design for quick iteration and disposable components.</li>
</ul>

<p><em>Originally published at <a href="https://www.technicalchops.com/articles/the-models-dont-matter-anymore">technicalchops.com</a>.</em></p>]]></content:encoded>
      <pubDate>Mon, 24 Feb 2025 00:00:00 GMT</pubDate>
      <author>simon@minton.co (Simon Minton)</author>
      <category>AI</category>
    </item>
    <item>
      <title>Chasing Cool: Jaguar’s Gamble</title>
      <link>https://www.technicalchops.com/articles/chasing-cool-jaguars-gamble</link>
      <guid isPermaLink="true">https://www.technicalchops.com/articles/chasing-cool-jaguars-gamble</guid>
      <description>Jaguar Redesign, EVs and China</description>
      <content:encoded><![CDATA[<p>Good morning,</p>
<p>Welcome to the first in a new series of Updates from Technical Chops. These updates will be available to all subscribers and follow the <a href="https://www.technicalchops.com/about/#schedule">schedule</a>.</p>


<p>Jaguar, the British luxury carmaker, unveiled a bold new brand identity last week and was promptly met with sharp criticism. Elon Musk, never one to pass up an opportunity to troll, summed up the sentiment with a single, cutting remark:</p>
<div class="flex justify-center">
<blockquote class="twitter-tweet" data-media-max-width="560"><p lang="en" dir="ltr">Do you sell cars?</p>— Elon Musk (@elonmusk) <a href="[https://twitter.com/elonmusk/status/1858914024013910020?ref_src=twsrc^tfw](https://twitter.com/elonmusk/status/1858914024013910020?ref_src=twsrc%5Etfw)">November 19, 2024</a></blockquote> <script async src="https://platform.twitter.com/widgets.js" charset="utf-8"></script>
</div>
<p>It’s fair to ask. Jaguar’s rebrand isn’t just aesthetic—it’s existential. On December 2nd, the company will reveal its first new model under the revamped identity, a GT concept, while simultaneously halting production and sales of its current lineup. Between now and the GT’s launch in 2026, Jaguar will have exactly zero cars for sale. As the company describes it, “this is a complete reset.”</p>

<p>This isn’t the kind of move you make when things are going well, but for Jaguar—an unprofitable brand struggling to define itself in a rapidly changing industry—it may be the only option left.</p>
<p>For decades, the automotive industry operated under rules that heavily favored incumbents. Scaling production required enormous amounts of capital and infrastructure, and a tangle of regulations acted as an effective moat. Over time, mergers and acquisitions further concentrated the market; today, just three companies—<em>Stellantis</em> (formerly <em>Chrysler</em>), <em>Volkswagen Group</em>, and <em>General Motors</em>—own 28 major car brands between them.</p>
<p>The rise of Electric Vehicles (EVs) has completely upended this dynamic. EVs are fundamentally different from internal combustion engine (ICE) vehicles. They’re simpler to manufacture, and the value of the product is shifting toward software, batteries, and sensors—areas where the traditional automakers are, at best, playing catch-up. Cars are becoming software-first consumer electronics—a category where China already excels.</p>
<p>And excel it has. China now has more car brands than the rest of the world combined, and its exports are booming. <a href="https://www.marklines.com/en/statistics/flash_prod/automotive-production-in-china-by-month">Chinese vehicle exports have risen 24% year-over-year so far in 2024</a>, following an extraordinary 58% increase in 2023. These aren’t just cheap knock-offs either; Chinese EVs combine high-quality design with competitive pricing, putting significant pressure on mid-market and low-luxury brands like Jaguar.</p>
<p>Traditional automakers face a serious challenge. Decades of safety regulations and converging design standards have made cars increasingly indistinguishable. So similar are today’s vehicles that it’s become a meme: try identifying a modern SUV from a silhouette alone. Meanwhile, in-car experiences are increasingly outsourced to Apple CarPlay and Android Auto, further diluting brand identity. Some automakers, like GM, are <a href="https://www.theverge.com/24285581/gm-software-baris-cetinok-apple-carplay-android-auto-google-cars-evs-decoder-podcast">attempting to regain control by ditching these platforms entirely</a>, but differentiation remains elusive.</p>
<p>One advantage legacy brands still hold is their history. Done right, nostalgia can be powerful. BMW turned the Mini into a modern success story, and Fiat built an entire lineup around the 500, spanning everything from sporty hatchbacks to compact SUVs. These brands don’t just sell cars—they sell a sense of continuity, a tie to something iconic, something with personality.</p>
<p>Jaguar, though, has struggled to harness its past. The E-Type—called “the most beautiful car in the world” by Enzo Ferrari—remains the brand’s most iconic achievement. But that was 60 years ago. Since then, Jaguar’s image has drifted. By the 1990s and 2000s, Jaguar had become the company car of choice for mid-level executives aspiring to a veneer of old-school prestige. These were cars that looked the part in the golf club car park but often came with a reputation for questionable reliability and a whiff of trying just a bit too hard.</p>
<p>Jaguar’s rebrand is an attempt to rewrite its narrative entirely. This isn’t about selling cars to everyone—it’s about selling Jaguar to the right people. The pivot to a fashion-forward identity is deliberate, aimed squarely at a younger, wealthier audience that browses <em>Vogue</em> rather than <em>Golf Digest</em>.</p>
<p>And here’s the thing: <em>Vogue</em> isn’t meant to be accessible. It’s aspirational by design. The E-Type didn’t just appear in <em>Vogue</em> because it was beautiful; it belonged there, side-by-side with haute couture. Jaguar’s new direction isn’t chasing mass-market appeal—it’s attempting to reclaim its place as a niche icon of cultural desirability.</p>
<p>The new branding, then, is a statement: Jaguar isn’t selling a car; it’s selling a lifestyle—rare, exclusive, and intentional. Like high fashion, it isn’t for everyone, nor does it aim to be. It’s a calculated bet that the brand can redefine itself for a market that values style and substance over accessibility.</p>
<p>For legacy automakers, the rise of high-quality, affordable Chinese EVs has made survival contingent on finding an identity that stands apart. Heritage can be a powerful tool, but only when it inspires rather than anchors. For Jaguar, the challenge isn’t just to recapture the allure of the E-Type or the glamour of the 1960s—it’s to reimagine what the brand could mean in a world that values reinvention as much as legacy.</p>
<p>Jaguar’s reinvention is polarizing by design—because it has to be. In an industry dominated by cautious evolution, it’s a rare, audacious gamble to rebuild a brand from the ground up. Whether this leap will redefine Jaguar as an icon for a new era—or consign it to nostalgia—will depend not just on what it builds, but on how much people still believe in the magic it once embodied.</p>


<ul>
<li>The threat from China’s EV manufacturers is made clear in <a href="https://x.com/TaylorOgan/status/1859146242519167249">this tweet</a> comparing one of Tesla’s Gigafactories to <a href="https://www.byd.com">BYD’s</a> Factory build-out in Zhengzhou. The company had capex of over $100bn as it has focussed on expansion over the past year, with <a href="https://www.bloomberg.com/news/articles/2024-10-30/byd-revenue-eclipses-tesla-for-first-time-as-ev-giants-go-head-to-head">revenue surpassing Tesla for the first time in Q3</a> of this year.</li>
<li>Staying in China, LLM Startup <a href="https://www.deepseek.com/">DeepSeek</a>, released a new model “DeepSeek-R1-Lite-Preview”. Aside from the fact that we need to get better at naming these things, the model has comparable performance to OpenAI’s o1-preview. With the US considering advanced AI to be a National Security concern, it is notable that DeepSeek is able to perform at frontier level, despite the <a href="https://www.bloomberg.com/professional/insights/technology/chinese-ai-firms-on-track-to-narrow-gap-despite-nvidia-controls/">US export controls that aim to stymie advancements outside of the US</a>.</li>
</ul>

<p><em>Originally published at <a href="https://www.technicalchops.com/articles/chasing-cool-jaguars-gamble">technicalchops.com</a>.</em></p>]]></content:encoded>
      <pubDate>Mon, 25 Nov 2024 00:00:00 GMT</pubDate>
      <author>simon@minton.co (Simon Minton)</author>
      <category>Auto</category>
      <category>Cars</category>
      <category>EVs</category>
      <category>China</category>
    </item>
    <item>
      <title>Balancing Speed and Accuracy in LLMs</title>
      <link>https://www.technicalchops.com/articles/balancing-speed-and-accuracy-in-llms</link>
      <guid isPermaLink="true">https://www.technicalchops.com/articles/balancing-speed-and-accuracy-in-llms</guid>
      <description>Speed, Accuracy, Considered Thought and Trust</description>
      <content:encoded><![CDATA[<p>Two weeks ago, I <a href="https://www.technicalchops.com/articles/consensus-in-the-machine/">explored a basic method for error-checking the output of Large Language Models (LLMs)</a>. Just a week later, OpenAI unveiled their latest model, o1 (Strawberry)—the first frontier model boasting built-in reasoning and iterative correction. This isn’t merely another version update; o1 signifies a fundamental shift in how we conceptualize reasoning within AI models and how we engage with them.</p>
<h3 id="the-evolution-of-ai-reasoning">The Evolution of AI Reasoning</h3>
<p>The arrival of o1 is exciting for two primary reasons. First, it incorporates a “Chain of Thought” mechanism, effectively deconstructing user prompts into logical steps to generate more accurate answers. Second, it utilizes this chain of thought to verify its own outputs—a more advanced iteration of the error-checking approach I discussed in my previous article.</p>
<p>But what does this mean in practical terms? <a href="https://www.oneusefulthing.org/p/something-new-on-openais-strawberry">Ethan Mollick’s crossword example</a> provides a clear illustration. When solving a crossword puzzle, multiple answers may fit a given clue, but only one will integrate with the rest of the puzzle. For instance, <em>A red or green fruit that is often enjoyed pressed as an alcoholic drink (5 letters)</em> could be either GRAPE or APPLE, but only one will align correctly with surrounding answers.</p>
<p>Most LLMs would generate GRAPE and move forward without re-evaluating its fit within the broader puzzle. In contrast, o1 excels by looping back, much like a human would, to recognize that GRAPE doesn’t fit and subsequently correct it to APPLE.</p>

<p>For the first time, AI responses undergo the “<em>that doesn’t look right, are you sure?</em>” <em>“Yes, sorry, you’re right, the correct answer is APPLE”</em> process, long familiar to regular LLM users, prior to being received by the user. Consequently, o1 and its successors are poised to deliver responses that are significantly more accurate than those of their predecessors.</p>
<h3 id="ais-inability-to-loop-back-a-one-shot-problem">AI’s Inability to Loop Back: A One-Shot Problem</h3>
<p>The limitation of previous models to revisit and rectify errors is a fundamental flaw in AI. Once a model generates its first token (word), it sets a trajectory that cannot be altered, even if the answer becomes evidently incorrect midway through. By the time an error is clear there’s no mechanism in place to correct it, leaving users with incorrect information.</p>
<p>There was, however, a brief period when AI appeared to be self-correcting, albeit not for accuracy. <a href="https://stratechery.com/2023/from-bing-to-sydney-search-as-distraction-sentient-ai/">When Ben Thompson described Sydney</a>—Microsoft’s “sentient” and unpredictable AI with a distinct personality—a phenomenon emerged where models would start an answer, delete it halfway, and revise it to comply with safety checks.</p>
<p>So, why hasn’t self-correction for accuracy been standard? The primary obstacles are the significant costs in processing power and time. For instance, o1’s output token cost is quadruple that of its predecessor, 4o, due to the increased number of tokens processed and generated. In Mollick’s crossword experiment, the self-correcting model required a full 108 seconds to solve the puzzle—substantially longer than typical models. Users generally resist slow and expensive systems, favoring the “one-shot” approach where LLMs provide immediate answers despite potential inaccuracies.</p>
<h3 id="instant-answers">Instant Answers</h3>
<p>Reflecting on computing history, early machines like Turing’s Bombe and <a href="https://en.wikipedia.org/wiki/Tommy_Flowers">Tommy Flowers’ Colossus</a> operated under the assumption that obtaining correct answers would take measurable processing time. The Bombe, for instance, took  20 minutes to crack an Enigma setting through sheer mechanical effort.</p>
<p>As technology advanced—silicon became faster, high-speed broadband became ubiquitous, and social media reshaped our expectations—we became accustomed to instant results. <a href="https://glinden.blogspot.com/2006/11/marissa-mayer-at-web-20.html">Marissa Mayer, as early as 2006,</a> championed the gospel of speed, emphasizing its importance: a mere half-second delay in loading Google results led to a 20% drop in traffic and revenue.</p>
<figure>
  <img src="https://www.technicalchops.com/images/ssagoogle.png" alt="Google Results Page">
</figure>
<p>For years, Google’s search results demonstrated the value placed on speed: ten blue links showcased alongside a proudly proclaimed search time of “<em>0.69 seconds</em>.” We’ve come to expect immediate gratification from the web, making the idea of waiting for an AI’s answer feel counterintuitive.</p>
<p>When OpenAI launched the ChatGPT interface, one of its standout features was the immediate, letter-by-letter streaming of responses—a user interface now standard across all LLMs. This technique creates an illusion of speed, keeping users engaged even as the AI continues to generate its final thoughts. However, this approach has a drawback: if the AI begins streaming responses before fully processing the answer, there’s no opportunity to correct errors that may emerge later in the generation process.</p>
<h3 id="thinking-fast-and-slow">Thinking, Fast and Slow</h3>
<p>Humans are often advised to take their time when responding to questions. Whether in interviews, tackling complex problems, or strategic planning, deliberate and reasoned thought typically yields better outcomes. Even individuals who seem to answer questions instantly have usually spent considerable time formulating their thoughts in advance.</p>
<p>Many readers will be familiar with <a href="https://en.wikipedia.org/wiki/Thinking,_Fast_and_Slow">Daniel Kahneman’s <em>Thinking, Fast and Slow</em></a>—the 2011 book that introduced his thesis of the dual systems of human cognition:</p>
<ul>
<li><strong>System 1:</strong> Handles fast, automatic, and unconscious thinking</li>
<li><strong>System 2:</strong> Manages slow, deliberate, and conscious reasoning</li>
</ul>
<p>Most current large language models (LLMs) operate primarily within the realm of System 1—they generate responses quickly and effortlessly, akin to off-the-cuff remarks.</p>
<p>Even with advancements like the o1 model, the process still resembles a System 1 response. Instead of engaging in true System 2 thinking, o1 appears to internally repeat and refine its initial System 1 output until it arrives at what it believes is the correct answer. This suggests that while o1 can self-correct, it’s still fundamentally reliant on rapid, surface-level processing rather than deep, reasoned thought.</p>
<p>One might argue that System 2 thinking in AI could manifest as error correction and analysis of System 1 outputs. However, the sheer volume of output tokens required indicates that the model is effectively “thinking out loud” internally, even if these thoughts aren’t explicitly shared with the user. This internal vocalization mirrors how humans might silently debate and refine their answers before speaking, but it doesn’t reflect the considered evolution of a thought over a significant period.</p>
<p>Considered thought is something that LLMs, as they currently operate, are simply not capable of. Even at the level of o1, they are trained to a cutoff point and don’t learn from interactions with users, nor do they have “defined” thoughts built up over time on any subject. <a href="https://www.youtube.com/watch?v=ahgjEjJkZks">The same question phrased in two different ways</a> could yield two vastly different responses because the AI is considering that question—and vitally its phrasing—for the first time, even on its ten-thousandth repetition.</p>
<h3 id="considered-thought">Considered Thought</h3>
<p>Could LLMs achieve considered thought akin to a human, though? This ventures into the realm of discussing Artificial General Intelligence (AGI)—a topic I prefer to sidestep for now. However, we must contemplate what future models might look like and whether they could be said to have “deliberated” over a topic.</p>
<p>A model that continuously learns from interactions and previous answers is the most likely candidate for exhibiting “considered thoughts.” The advent of such a model, however, remains distant, as <a href="https://arxiv.org/abs/2402.01364">it is prohibitively expensive in terms of time and compute</a> (and, undoubtedly, financial resources) to regularly, let alone continuously, retrain a model. Nevertheless, it seems likely that this will be the eventual evolution of these models, with the ability to learn from all past interactions and build upon previous answers.</p>
<p>Until then, we’re likely to be left with models that use processes like o1 to imitate reasoned thought, checking and refining their responses before delivering them to the user.</p>
<h3 id="balancing-user-expectations-of-instantaneity-with-correct-answers">Balancing User Expectations of Instantaneity with Correct Answers</h3>
<p>Herein lies a profound challenge: users expect instant, accurate results, but by their very nature, LLMs achieve their highest accuracy when they don’t respond immediately. How do we reconcile these conflicting expectations?</p>
<p>We could allocate more processing power to make the system appear faster, but the cost implications are substantial. Alternatively, can we educate users to understand that AI requires “thinking time”? While instant responses align with user expectations shaped by the internet’s speed, there’s a growing imperative to prioritize accuracy and depth of reasoning to ensure that the responses delivered by LLMs are trustworthy and dependable.</p>
<h3 id="the-dawn-of-trustworthy-ai">The Dawn of Trustworthy AI</h3>
<p>o1;s capacity for reasoned thought and self-correction goes beyond merely improving the outputs previous models—it marks a significant stride toward creating genuinely intelligent systems capable of partnering with us in addressing the complex challenges of our world.</p>
<p>As AI models advance towards incorporating internal reasoning and self-correction, we need to consider how we can manage user expectations and ensure they understand that the correct answer isn’t necessarily the immediate one. Balancing the innate human desire for instant results with the necessity for accurate, thoughtful responses will be paramount. By valuing deliberative processes alongside speed, we will start to see AIs applications that can be trusted to actually deliver on AI’s early promise.</p>


<p>The current generation of LLMs works well in low-stakes environments where high trust isn’t essential, but they still need significant human oversight. Models like o1 change that. With their ability to handle more complex tasks and self-correct, these models open the door to trusting AI in critical roles. This shift could save companies money, reduce errors, and unlock new possibilities by allowing AI to take on more responsibility and deliver reliable results in challenging scenarios.</p>


<ul>
<li><strong>A Paradigm Shift in AI Reasoning</strong>: OpenAI’s o1 (Strawberry) model introduces built-in reasoning and self-correction, moving beyond the “one-shot” approach to improve accuracy.</li>
<li><strong>Self-Correction in Action</strong>: Unlike its predecessors, o1 employs a “Chain of Thought” mechanism to iteratively verify and refine its responses before presenting them to the user.</li>
<li><strong>Accuracy vs. Speed Trade-offs</strong>: While o1 achieves higher accuracy, its self-correction process significantly increases the computational cost and response time, posing challenges for user expectations of instant results.</li>
<li><strong>Toward Trustworthy AI</strong>: The evolution of models like o1 highlights a critical shift toward trustworthy, thoughtful AI—though true considered thought and learning from interactions remain distant goals.</li>
<li><strong>Reconciling Expectations</strong>: Balancing users’ demand for immediacy with the benefits of deliberate, accurate reasoning will be central to the future of AI-human interactions.</li>
</ul>

<p><em>Originally published at <a href="https://www.technicalchops.com/articles/balancing-speed-and-accuracy-in-llms">technicalchops.com</a>.</em></p>]]></content:encoded>
      <pubDate>Sun, 22 Sep 2024 00:00:00 GMT</pubDate>
      <author>simon@minton.co (Simon Minton)</author>
      <category>AI</category>
      <category>Technical</category>
      <category>LLMs</category>
    </item>
    <item>
      <title>The Future of Mobile: Contextual Computing, App Clips, and Apple’s Privacy Revolution</title>
      <link>https://www.technicalchops.com/articles/the-future-of-mobile-contextual-computing</link>
      <guid isPermaLink="true">https://www.technicalchops.com/articles/the-future-of-mobile-contextual-computing</guid>
      <description>Contextual computing is reshaping how we interact with technology.</description>
      <content:encoded><![CDATA[<p><em>This article has a strong Apple-focus. I’m aware that similar systems exist within Android, I’m simply most familiar with Apple and generally try to stick to what I know.</em></p>
<p>Prior to 2020, the QR code seemed destined to live perpetually in the “almost made it” category of tech innovations. Despite its widespread use in industry, spotting one on a poster or product often left me wondering how it had survived the design process without someone suggesting, <em>“Perhaps we should just make this easier to Google?”</em></p>
<p>And then, COVID happened.</p>
<p>The QR code emerged as an unlikely winner of the pandemic. Suddenly, these pixelated squares became portals to test entry, contact tracing, emergency information, and restaurant menus. QR codes transformed from an ugly design element slapped on the corner of a poster to a central feature of our daily lives.</p>
<figure>
  <img src="https://www.technicalchops.com/images/qr-codes.png" alt="Pandemic-era QR codes">
  <figcaption>Pandemic-era QR codes</figcaption>
</figure>
<p>This shift revealed two key strengths of QR codes:</p>
<ul>
<li>They provided users with effortless access to digital experiences.</li>
<li>They supplied apps with contextual information about the user without requiring manual input.</li>
</ul>
<p>In essence, QR codes enabled a user experience that designers had long envisioned: the shortest possible path between user intent and goal completion. Without realising it, we were experiencing the introduction of a type of experience that will come to define how we interact with our devices: Contextual Computing.</p>
<h3 id="app-clips">App Clips</h3>
<p>In the midst of the 2020 QR code renaissance, Apple introduced App Clips. These are small, install-free portions of apps that offer specific functionality, “discoverable at the moment it’s needed.” While the timing seemed opportunistic, given the newfound appreciation for QR codes, App Clips represented a more strategic move by Apple.</p>
<figure>
  <img src="https://www.technicalchops.com/images/app-clips.png" alt="the App Clip is invoked by scanning a code, allowing the user to order and pay for a donut.">
  <figcaption>In the example from [Apple’s documentation](https://developer.apple.com/documentation/appclip) above, the App Clip is invoked by scanning a code, allowing the user to order and pay for a donut.</figcaption>
</figure>
<p>App Clips address a growing issue in the current app ecosystem: people don’t like installing apps anymore. When the App Store launched in 2008, it was accompanied by the brilliant tagline “<em>There’s An App For That</em>.” Apple quickly realized that the ecosystem of apps would be vital to the iPhone’s success. Today, there are roughly 1.8 million apps in the store. And yet, I find myself using fewer and fewer apps.</p>
<p>It’s not just me. <a href="https://www.businessofapps.com/news/deciphering-the-app-store-slowdown-why-are-installs-declining">Research shows that app install rates are falling</a>, with users tending to have fewer apps installed than they did just a couple of years ago. There are many potential reasons for this. “App Fatigue” has led users to avoid installing additional apps, viewing them as just more overhead. The barrage of push notifications, relevant or not, has fostered frustration, especially when they come from long-dormant apps. Many users are particularly annoyed by apps they’re required to install for a single event, only to delete the next day.</p>
<p>But perhaps the primary reason is that the process of installing an app is often convoluted for the specific tasks users are trying to accomplish, especially for one-off situations.</p>
<p>Take paying for parking: where once you walked up to a machine, inserted a coin, and pressed a button to print your ticket, now you must download an app, sign up, confirm your email, add your vehicle details, type in the parking location code, potentially add credit card details if Apple Pay isn’t supported, and then, finally, pay for parking. It’s hardly an improved experience.</p>
<h3 id="the-data-collection-conundrum">The Data Collection Conundrum</h3>
<p>Part of the reason apps have these convoluted processes is the insatiable desire to “own” customer data. For most companies, it’s simply not enough for the customer to use and pay for their service.</p>
<p>In rare cases, this extensive data collection might be justified. But for most, it’s unnecessary. The parking company, for instance, needs at most the license plate and parking location to ensure the payment is associated with the right vehicle. An email for the receipt? That’s already provided by a proper implementation of the payment system. There’s no compelling reason (beyond marketing) to require users to set up an account<sup><a href="#user-content-fn-1" id="user-content-fnref-1" data-footnote-ref="" aria-describedby="footnote-label">1</a></sup>.</p>
<p>It often feels as though many apps have forgotten that their primary goal is to deliver a service or product that users will pay for. In the case of ride-sharing or food delivery apps, it sometimes seems their main purpose is to send marketing notifications encouraging use of their services… which ironically now puts me off using them.</p>
<h3 id="apples-push-back">Apple’s Push Back</h3>
<p>In recent years, Apple has made it abundantly clear that user privacy is a top priority. The introduction of “Sign in with Apple” in 2019 was a significant step in this direction. What initially appeared to be just another sign-in option alongside Facebook and Google logins is actually part of Apple’s broader strategy to protect user privacy.</p>
<p>The system uses a “Private Relay” email address by default, providing apps with a single-use email that redirects to the user’s actual address. This limits the information shared with the app provider - a data loss for the app, perhaps, but a privacy win for the user.</p>
<p>Apple’s requirement that Sign in with Apple must be implemented if other third-party logins are offered may frustrate developers, but it’s a boon for users. Sign in with Apple didn’t exist at the start of 2019, but was ubiquitous by year’s end.</p>
<p>From the effect of <a href="https://www.vox.com/recode/23045136/apple-app-tracking-transparency-privacy-ads">Ad Tracking Transparency</a> to <a href="https://www.litmus.com/blog/apple-mail-privacy-protection-for-marketers">Mail Privacy Protection</a>, it’s clear that Apple prioritizes protecting users over improving marketing efficacy. To some extent, it feels as though Apple is rectifying the missteps made in the early App Store years.</p>
<h3 id="the-promise-of-app-clips">The Promise of App Clips</h3>
<p>Given Apple’s reflection on past mistakes and subsequent corrective actions, it’s worth considering their motivations for creating App Clips. The limitations imposed on App Clips suggest they’re part of Apple’s broader strategy to enhance user experience while maintaining privacy:</p>
<ul>
<li>Fast, install-free experience</li>
<li>Limited push notifications (8 hours after each launch)</li>
<li>Login restricted to Sign in with Apple</li>
<li>Focused on a single experience per App Clip</li>
<li>Payments via Apple Pay only</li>
<li>Automatic deletion after 30 days of disuse</li>
</ul>
<p>These user-centric choices may not align with developers whose primary goal is maximizing app installs. However, they cater perfectly to developers focused on delivering the service behind the app.</p>
<h3 id="apple-intelligence-app-clips-and-actions">Apple Intelligence, App Clips and Actions</h3>
<p>When I wrote about <a href="https://www.technicalchops.com/articles/apple-and-ai/">Apple and AI</a> back in February, I talked about what I’d hoped for from Apple Intelligence. One of the wishes that I had was that we had <em>“Fewer Apps, more Actions”.</em></p>
<p>While App Clips were introduced in 2020, their full potential may be realized through integration with Apple’s AI initiatives. As Apple Intelligence gains greater contextual awareness through access to emails, messages, and location data, it can surface relevant App Clips at precisely the right moment.</p>
<figure>
  <img src="https://www.technicalchops.com/images/phone-booking.png" alt="A Siri Suggestion to book a restaurant appears in a message thread.">
</figure>
<p>I want Apple to use its knowledge of the actions available within Apps–even ones that I don’t have installed–to allow me to achieve my goals easily. For example, if I’m messaging a friend about dinner plans, an incredible experience would be being offered access to an App Clip which allows me to make a reservation right from my conversation.</p>
<p>To me, these seamless, contextually aware experiences represent the future of mobile computing.</p>
<p>For the companies that produce apps, embracing this new paradigm means focusing on creating well-designed App Clips and implementing “Actions” that can be triggered by Apple Intelligence based on user context. By shifting focus from “building an app” to “delivering a service,” developers can create more valuable, user-centric experiences that align with the evolving landscape of mobile computing.</p>
<h3 id="refocusing-for-the-era-of-contextual-computing">Refocusing for the Era of Contextual Computing</h3>
<p>When we look at Apple’s overall approach–the focus on user privacy, the increasing prominence of actions in the OS, the release of App Clips–a core message emerges. The shift towards contextual computing will require that companies refocus on their primary value proposition. The era of building an app for the sake of having an app presence is coming to an end.</p>
<p>Instead, businesses should ask themselves:</p>
<ol>
<li>What core service or product are we providing?</li>
<li>How can we deliver this value to users in the most frictionless way possible?</li>
<li>What are the key moments when our service is most needed, and how can we be present in those moments?</li>
</ol>
<p>By centering on these questions, companies can move away from the mindset of “We need an app” to “We need to be available when and where our customers need us.” This shift aligns perfectly with the promise of App Clips and contextual computing.</p>
<p>For developers, this paradigm shift presents both challenges and opportunities:</p>
<ol>
<li><strong>Rethinking User Engagement</strong>: Instead of focusing on metrics like daily active users or time spent in app, developers will need to optimize for successful task completion and user satisfaction in brief, contextual interactions.</li>
<li><strong>Modular Design</strong>: Creating effective App Clips requires a modular approach to app design, where core functionalities can stand alone and be easily surfaced in different contexts.</li>
<li><strong>Context-Aware APIs</strong>: Developers will need to become adept at leveraging the new APIs and frameworks that allow their services to be discoverable and actionable in various contexts.</li>
<li><strong>Privacy-First Development</strong>: With Apple’s strong stance on privacy, developers must find ways to deliver personalized, contextual experiences while minimizing data collection and storage.</li>
</ol>
<h3 id="embracing-the-future-of-contextual-computing">Embracing the Future of Contextual Computing</h3>
<p>While the journey towards truly contextual computing is just beginning, it promises to fundamentally reshape how we interact with digital services. By focusing on delivering core value in the moments that matter most, companies can position themselves to thrive in this new paradigm, creating experiences that are both more useful and less intrusive in our daily lives.</p>
<p>As we move into this new era of contextual computing, the rules of engagement are shifting. The focus is no longer on how many apps users install, but on delivering meaningful experiences at the right moment—when they are most relevant and needed. For businesses and developers, this transition presents both a challenge and a tremendous opportunity. Those who adapt to this new paradigm, prioritizing frictionless user interactions and respecting privacy, will be the ones who thrive.</p>
<p>Apple’s vision reflects a broader shift toward user-centric, privacy-first computing. By reducing the clutter of endless app installs and putting seamless functionality at the forefront, Apple is leading the way in transforming how we interact with digital services.</p>
<p>The future of mobile technology isn’t about building apps for every conceivable use case—it’s about intelligently delivering actions and services when and where they’re needed most.</p>
<p>In the end, those who embrace this shift will not only improve their user experience but will also help define the next generation of digital interactions. The age of contextual computing is here, and it’s time for companies to focus on what matters most: being present and purposeful in the lives of their users.</p>
<section data-footnotes="" class="footnotes"><h2 class="sr-only" id="footnote-label">Footnotes</h2>
<ol>
<li id="user-content-fn-1">
<p>The counter argument to this is that the parking app stores a history (and present state) or a given vehicle’s parking which could lead to privacy risk if this was accessible only via license plate. The obvious answer to this is to associate that history/state with a given device only. We got on pretty well for many decades without needing to pull up our parking histories. <a href="#user-content-fnref-1" data-footnote-backref="" class="data-footnote-backref" aria-label="Back to content">↩</a></p>
</li>
</ol>
</section>
<p><em>Originally published at <a href="https://www.technicalchops.com/articles/the-future-of-mobile-contextual-computing">technicalchops.com</a>.</em></p>]]></content:encoded>
      <pubDate>Tue, 17 Sep 2024 00:00:00 GMT</pubDate>
      <author>simon@minton.co (Simon Minton)</author>
      <category>AI</category>
      <category>LLMs</category>
      <category>Mobile</category>
    </item>
    <item>
      <title>Consensus in the Machine: A New Method for Tackling LLM Hallucinations</title>
      <link>https://www.technicalchops.com/articles/consensus-in-the-machine</link>
      <guid isPermaLink="true">https://www.technicalchops.com/articles/consensus-in-the-machine</guid>
      <description>Triple Modular Redundancy for LLMs</description>
      <content:encoded><![CDATA[<p><em>Programming note: Technical Chops is back from the August break. From this week on, posts will be delivered on a Friday morning.</em></p>
<p>One of the challenges with Large Language Models (LLMs) is that they are still, quite often, incorrect. Worse still, they are confidently incorrect. They are, to put it politely, bullshitters.</p>
<h3 id="software-in-critical-environments">Software in Critical Environments</h3>
<p>Within a critical environment, we build in a way which assumes that software, or the hardware that runs or is an input to that software, can be wrong.</p>
<p>Airliners are an acute example – they don’t just have one sensor for a given metric, they often have at least three. By using the data from three sensors, their software can use a methodology called Triple Modular Redundancy (TMR) to identify when one of the sensors is providing bad data. If one airspeed sensor starts saying that the plane is going too slowly, but the other two sensors say everything is normal, TMR allows the plane to effectively take the “votes” of the two normal sensors as the correct speed. This is especially critical when the plane can take actions by itself based on those inputs; the lack of it was a <a href="https://perell.com/essay/boeing-737-max/">contributory factor to the Boeing 737 Max disasters in 2018 and 2019</a>.</p>
<figure>
  <img src="https://www.technicalchops.com/images/airspeed.png" alt="An illustration of three airspeed indicators where one is not in agreement with the other two.">
  <figcaption>Basic TMR for an Airline (📷 Pascal Borener)</figcaption>
</figure>
<p>Thankfully, I can’t think of any good uses for LLMs in critical environments. Whilst there have been cases where <a href="https://reason.com/volokh/2023/05/27/a-lawyers-filing-is-replete-with-citations-to-non-existent-cases-thanks-chatgpt/">lawyers have cited fake cases created by ChatGPT</a> and faced the resultant wrath of a judge, most of the time the stakes are your own, or your product’s, credibility.</p>
<p>Most people like protecting their credibility though, and so there are use cases which would be perfect for an LLM but which we tend to avoid implementing because of the risk of the AI hallucinating an answer.</p>
<h3 id="learning-from-critical-environments">Learning from critical environments</h3>
<p>We tend not to implement TMR in environments where it’s not strictly necessary. It’s an unnecessary overhead, in terms of maintenance, complexity and of course cost. Most of the time, properly written software will do exactly what it is supposed to do, with <a href="https://web.mit.edu/jemorris/humor/500-miles">bugs and edge cases more of a curiosity</a> that are easily solved once identified.</p>
<p>The issue with LLM hallucinations though is that we can’t “solve” them. And they happen inconsistently, precisely because of the way that LLMs work. If you ask the same LLM the same question phrased in slightly different ways, you will still get <a href="https://chatgpt.com/c/9635589d-1eb4-45fa-9812-93061d0e0600">varying</a> <a href="https://chatgpt.com/c/65b3939d-4a79-434c-822d-9a308615222b">answers</a>.</p>
<p>One way we try to get better results from a model is by “fine-tuning” a model. This means that a model is trained on specific content or a use-case so that it better understands what it needs to do. This is quite often costly, but it does work well, right up until the point that a new frontier model is produced that outperforms the fine-tuned model. At that point the investment made into the fine-tuned model is nullified because it makes more sense to switch to the outperforming model.</p>
<h3 id="making-use-of-tmr-in-llms">Making use of TMR in LLMs</h3>
<p>The cost of LLMs has plummeted over the past two years. Standing at around $2 per 100,000 words<sup><a href="#user-content-fn-1" id="user-content-fnref-1" data-footnote-ref="" aria-describedby="footnote-label">1</a></sup>, we’re now in the realm that for any decent business use-case, model costs aren’t a huge consideration.</p>
<p>What if we used three (or more) different models simultaneously and provided them with the same query, and then allowed them to vote on the best answer? Perhaps implementing a form of TMR could solve the problem of inaccuracy within LLMs.</p>
<p>Well, the only way to find out is to actually do it: <a href="https://consensus.technicalchops.com/basic">https://consensus.technicalchops.com/basic</a></p>
<p>I’ve kept it simple, but the concept is pretty straight forward: Ask a question and GPT-4o, Claude 3.5 Sonnet and Google Gemini 1.5 Pro will all provide an answer. Those answers are then fed back into the same models (without previous context) and the models are asked to vote for the best answer.</p>
<p>We can take this a step further though. What if we allow the models to correct each other’s work prior to choosing the best optimised answer? The results is that we do tend to get better answers. You can give the optimised version a go here: <a href="https://consensus.technicalchops.com/optimise">https://consensus.technicalchops.com/optimise</a></p>
<p>The delay in the response is noticeable, but I think the idea has the possibility to improve accuracy when using off the shelf LLMs<sup><a href="#user-content-fn-2" id="user-content-fnref-2" data-footnote-ref="" aria-describedby="footnote-label">2</a></sup> and, vitally, to catch the mistakes that cause us to distrust them.</p>
<p>The code is available here: <a href="https://github.com/simonminton/consensus-article">https://github.com/simonminton/consensus-article</a></p>
<section data-footnotes="" class="footnotes"><h2 class="sr-only" id="footnote-label">Footnotes</h2>
<ol>
<li id="user-content-fn-1">
<p>GPT4o-mini is $5/1M input tokens and $15/1M output <a href="#user-content-fnref-1" data-footnote-backref="" class="data-footnote-backref" aria-label="Back to content">↩</a></p>
</li>
<li id="user-content-fn-2">
<p>I’m aware of ensemble methods within ML - the focus here is on off the shelf LLMs <a href="#user-content-fnref-2" data-footnote-backref="" class="data-footnote-backref" aria-label="Back to content">↩</a></p>
</li>
</ol>
</section>
<p><em>Originally published at <a href="https://www.technicalchops.com/articles/consensus-in-the-machine">technicalchops.com</a>.</em></p>]]></content:encoded>
      <pubDate>Thu, 05 Sep 2024 00:00:00 GMT</pubDate>
      <author>simon@minton.co (Simon Minton)</author>
      <category>AI</category>
      <category>Technical</category>
      <category>LLMs</category>
    </item>
    <item>
      <title>AI: The Ultimate Sherlocking?</title>
      <link>https://www.technicalchops.com/articles/ai-the-ultimate-sherlocking</link>
      <guid isPermaLink="true">https://www.technicalchops.com/articles/ai-the-ultimate-sherlocking</guid>
      <description>History Doesn&apos;t Repeat Itself, but It Often Rhymes</description>
      <content:encoded><![CDATA[<p>If you’re familiar with the Apple App world you may sometimes have heard someone refer to an app having been “Sherlocked”. This basically means that Apple has decided to add that App’s core feature to their main operating system, be it iOS, MacOS or iPadOS. It does not usually end well for the original app.</p>
<h3 id="sherlocked">Sherlocked?</h3>
<p>Way back, in the mists of time (1998), Mac OS came with an app called Sherlock. It was originally an extension to Finder that allowed users to search not just by file name, but also within the contents of files.</p>
<p>Sherlock did more than that though. It also allowed users to search on the Internet, without using a browser. You could tap in, say, “Sherlock Holmes” and select multiple search engines — AltaVista, Excite and Lycos of course — and it would pull up the results. This was quite a cool thing to be able to do.</p>
<p>Sherlock became very popular and the next version of the software let users use the same system-based, universal interface to search the web via different channels — Entertainment, Pictures, Shopping, News — for whatever the user wanted. An entrepreneurial company then built another piece of software: Watson. Watson was similar to Sherlock, but it allowed you to search on specific sites; ebay, Amazon and more. Not only that, but it also allowed you to buy things on these sites—it was sort of the first universal e-commerce app. This was a very cool thing to be able to do!</p>
<p>Apple, keen to improve its own tool, then added to Sherlock the ability to… search on specific sites; ebay, Amazon and more. Not only that, but it also allowed you to buy things on these sites. Ah.</p>
<p>Many websites rushed to develop their own channels or extensions for Sherlock so that they could appear within the app. Sherlock was a success and Apple was very happy.</p>
<p>The makers of Watson though, were not happy. Apple, at a time when they still responded to these things, argued that the features were an inevitable evolution of Sherlock. Watson thus became the first app to be “Sherlocked”.</p>
<p>Well, that’s a relatively neat origin story. The pattern remains the same to this day. Every year, Apple improves their own software by adding features that until that point third-party apps have solved, and websites like <a href="https://techcrunch.com/2023/06/12/all-the-things-apple-sherlocked-at-wwdc-2023">TechCrunch get a good annual article to write</a>.</p>
<p>Sherlock the app though, eventually disappeared, tumbling to its death in 2007.</p>
<h3 id="theres-an-app-for-that">There’s An App For That</h3>
<p>At around the same time that Sherlock disappeared, the tech industry and the businesses which came to dominate the following decade, were starting to understand the importance of analytics data and the role that their sites’ designs played in both user retention and engagement. As their strategies improved, many sites closed down avenues through which users could engage via third parties like Sherlock. It became increasingly important to “own” the relationship with the consumer.</p>
<p>The arrival of Apps in 2008 saw companies realise that these offered an even more direct means of owning that relationship. We saw websites push users towards installing apps on their phones, partially so that there was an ever-present reminder of what to tap on to find what you’re looking for but also to gain access to those all-important notifications. Third party tools like Sherlock were no more, but now we lived in the future, where we could wake up in the middle of the night to our phones buzzing with an important offer on socks.</p>
<p>We also saw the emergence of the “big players”.</p>
<p>Sites like Amazon had realised that the real play wasn’t just in being a direct seller, it was in being the aggregator of <em>all</em> sellers. For Amazon, this, in combination with Bezos’ maniacal focus on customer experience, led to the Amazon we know today: a massive and comprehensive product inventory, free next-day shipping and one-click checkout. The Amazon experience is so well executed that half of all users <a href="https://www.notion.so/AI-The-Ultimate-Sherlocking-9dbcebf3aae94553ae849e14fddb666e?pvs=21">don’t start searches for products on Google, they go directly to Amazon</a>.</p>
<p>And this has been the status quo for many years now. We have a handful of incredibly well executed aggregators that own the various different sectors. Aggregators of this sort, <a href="https://stratechery.com/aggregation-theory/">as you may be familiar</a>, dominate the internet. All of the biggest players in every space are in some form or another, a type of aggregator — Amazon for retail, booking.com for holidays, Netflix for entertainment, uber for taxis — these businesses don’t necessarily provide the end product from their own supply, but they do provide access to the service <strong>in the most convenient way.</strong></p>
<p>The key for aggregators is their direct relationship with the user, which Ben Thompson describes as <a href="https://stratechery.com/2017/defining-aggregators/"><em>“the linchpin on which everything else rests”</em></a>. As long as the aggregators can maintain this direct relationship and ensure that their service provides the most convenient way for users to access the services, they will maintain their dominance.</p>
<h3 id="ai-personal-assistants">AI Personal Assistants</h3>
<p>In January, I wrote about <a href="https://www.technicalchops.com/articles/how-hard-is-being-an-assistant-anyway/">AI based Personal Assistants</a>, sparked by emerging products from companies like Humane and Rabbit. Those hardware devices, as I’d suspected, were DOA, with MKBHD describing the Humane Pin as <a href="https://www.youtube.com/watch?v=TitZV6k8zfA">“The Worst Product I’ve Ever Reviewed”</a>. But does this mean the idea of an AI personal assistant is dead? Not quite.</p>
<p>The reason that humane and rabbit stood out was because they were a new class of hardware; if they’d just been apps then they’d have received relatively little airtime. And it turned out that <a href="https://www.androidauthority.com/rabbit-r1-is-an-android-app-3438805/">rabbit was just an app</a>, in a fancy new housing. The problem however with being “just an app” goes beyond marketability.</p>
<p>The challenge with creating an effective AI personal assistant is access. It needs to interact with everything on your device. This level of integration is only truly possible for device manufacturers like Apple, Google, and Samsung. Apple, in particular, seems to be embracing this opportunity with Apple Intelligence and Siri.</p>
<p>While headline AI features baked into the OS—improved contextual search, writing tools, smart replies, summarized notifications—are impressive, Siri’s real power as an AI PA lies in its ability to take actions within other apps, potentially without direct user interaction.</p>
<p>At first glance, this seems like a win for everyone. Imagine Siri popping up with a timely notification to book an Uber when it recognizes you have a dinner reservation in 20 minutes. It’s like ultra-targeted advertising for free. But there’s a catch: Uber (and companies like it) risk losing their direct relationship with users.</p>
<h3 id="the-aggregators-dilemma">The Aggregator’s Dilemma</h3>
<p>This puts these businesses in a tricky spot. Users adapt quickly to new paradigms. If it becomes normal to rely on Siri (or its Android equivalent) as the first port of call for actions like booking a car or a table, apps will need to provide these actions to Siri or watch as competitors do so. However, this loss of direct interaction means services become more interchangeable, with the customer relationship shifting from the app to Siri. This is the risk that I wrote about in <a href="https://www.technicalchops.com/articles/ai-as-a-personal-aggregator/">AI as a Personal Aggregator</a>.</p>
<p>The big aggregators stand to have the linchpin Thompson talks about snapped – if the first place that a user goes to find what they need is Siri, then they lose one of the structural pillars on which their market dominance rests.</p>
<p>This shift towards AI-driven personal assistants as primary interfaces carries profound implications for the tech ecosystem. For users, it promises unprecedented convenience—imagine a world where your device anticipates your needs, seamlessly integrating various services without you having to juggle multiple apps. Need a ride? Siri books it. Forgot to order groceries? Your AI assistant has already compiled a list based on your usual purchases and dietary preferences.</p>
<p>However, for app developers and service providers, this new paradigm presents both opportunities and challenges. On one hand, integration with AI assistants could provide access to a broader user base and more contextual usage. On the other, it risks commoditising their services and weakening brand loyalty.</p>
<p>For the tech giants developing these AI assistants, it’s a potential goldmine. By positioning themselves as the gatekeepers of user interaction, they could gain unprecedented insight into user behavior and preferences. This data could be leveraged to further improve their services, creating a virtuous cycle of increasing utility but also dependence.</p>
<h3 id="the-final-chapter">The Final Chapter</h3>
<p>Of course, the aggregators won’t go down without a fight. They’ll integrate, they’ll innovate, they’ll probably coin some buzzwords about “AI-enhanced user experiences.” But at the end of the day, they might just be delaying the inevitable. After all, why juggle a dozen apps when one AI assistant can do it all for you?</p>
<p>In a twist of digital déjà vu, we seem to be circling back to a universal interface for accessing third-party services on our devices. Perhaps Sherlock wasn’t dead at all, maybe it just has a new name: Siri.</p>
<p><em>Originally published at <a href="https://www.technicalchops.com/articles/ai-the-ultimate-sherlocking">technicalchops.com</a>.</em></p>]]></content:encoded>
      <pubDate>Fri, 26 Jul 2024 00:00:00 GMT</pubDate>
      <author>simon@minton.co (Simon Minton)</author>
      <category>AI</category>
      <category>Assistants</category>
      <category>Agents</category>
      <category>Mobile</category>
      <category>Long Term Effects</category>
      <category>LLMs</category>
      <category>Apple</category>
    </item>
    <item>
      <title>Wu-Tang Forever? The Bizarre Afterlife of a $4 Million Album</title>
      <link>https://www.technicalchops.com/articles/wu-tang-forever</link>
      <guid isPermaLink="true">https://www.technicalchops.com/articles/wu-tang-forever</guid>
      <description>Once Upon a Time in DAOlin</description>
      <content:encoded><![CDATA[<p>Let’s take a short break from AI and hit a topic I have purposely avoided until now: Crypto.</p>
<h3 id="the-wu-tang-clan">The Wu-Tang Clan</h3>
<p>Wu-Tang Clan provided much of the soundtrack to my teenage years; their unique, raw <a href="https://daily.redbullmusicacademy.com/2018/06/engineering-wu-tang-clan">approach to production</a> and <a href="https://pudding.cool/projects/vocabulary/index.html">writing ability</a> stood out amongst the ocean of gangsta rap that flooded the early 2000s. Still now, if I need to get stuff done, I stick them on. In fact, 36 Chambers is playing as I write this. It would be fair to say that I am a fan.</p>
<p>From <a href="https://www.TheAlbum.com">TheAlbum.com</a>:</p>
<blockquote>
<p>Ten years ago, the visionary Wu-Tang Clan released an album unlike any other in the world. Once Upon a Time in Shaolin was recorded in secret over six years and, as a protest against the devaluation of music in the digital era, only one physical copy of the album was created. Once Upon a Time in Shaolin went on to become the most expensive work of music ever sold — twice.</p>
</blockquote>
<p>Basically, the Wu-Tang Clan made an album, pressed only one copy and sold it to one single buyer for a couple of million dollars. The purchase included an agreement that prevented the owner from selling copies of it for 88 years, although they were allowed to release it for free or play it at listening parties. The idea, according the Wu-Tang was:</p>
<blockquote>
<p>The music industry is in crisis. The intrinsic value of music has been reduced to zero. Contemporary art is worth millions by virtue of its exclusivity … By adopting a 400 year old Renaissance-style approach to music, offering it as a commissioned commodity and allowing it to take a similar trajectory from creation to exhibition to sale … we hope to inspire and intensify urgent debates about the future of music.</p>
</blockquote>
<p>That sounds fair enough; the album as a work of art, intrinsically valuable because of its exclusivity. There was always an element of risk of course; unlike a sculpture or painting, the value isn’t in the original physical manifestation of the object, it’s in the fact that the only way to hear the music is to have the original copy …unless someone were to… make a copy of it.</p>
<h3 id="pharma-bro">Pharma Bro</h3>
<p>So Wu Tang put One Upon a Time in Shaolin up for auction. Inevitably, a perfectly normal and reasonable art collector ends up as the owner: Martin Shkreli. You are likely somewhat familiar with Martin, the internet villain and <a href="https://www.theguardian.com/us-news/2024/jan/24/martin-shkreli-drug-industry-ban">“pharma bro” entrepreneur who increased the price of a life-saving drug by an astronomic 4,000%</a>. It cost him a cool $2m in 2015 and for the most part, he kept the music away from prying ears, save for a few minutes shared on live streams and listening parties. When <a href="https://www.justice.gov/usao-edny/pr/martin-shkreli-sentenced-seven-years-imprisonment-multi-million-dollar-fraud-scheme">Shkreli was, surprisingly, arrested later in 2015 and subsequently convicted in 2018</a>, the Department of Justice went after his assets and seized the album to pay for his debts.</p>
<p>The DoJ, not known for their great love of rappers (again, my teen years led me to understand that the feeling is mutual), was sadly not willing to <a href="https://x.com/matt_levine/status/677520311470989313">“<em>release it for free as part of our cultural patrimony</em>”</a> and sought a buyer, eventually selling it for ~$4m in 2021. That’s a pretty good return for Martin, even if he didn’t get to see the actual cash, and of course, was never able to listen to the album again. The Album, by this point having acquired enough cultural legacy, then of course ended up in the hands of a totally normal, straightforward billionaire and went on to spend the the next 80 years locked in a vault. Right?</p>
<h3 id="daos">DAOs</h3>
<p>Obviously not. The eventual owner, after a convoluted process to allow them to purchase it using cryptocurrency, was <a href="https://www.nytimes.com/2021/10/20/arts/music/wu-tang-clan-once-upon-a-time-in-shaolin.html">PleasrDAO</a>. PleasrDAO isn’t a person or company<sup><a href="#user-content-fn-1" id="user-content-fnref-1" data-footnote-ref="" aria-describedby="footnote-label">1</a></sup> but a DAO; according to <a href="https://www.investopedia.com/tech/what-dao/">Investopedia</a>:</p>
<blockquote>
<p>A decentralized autonomous organization (DAO) is an emerging form of organizational structure with no central governing body and whose members share a common goal of acting in the best interest of the entity. Popularized by blockchain enthusiasts, DAOs make decisions using a bottom-up management approach.</p>
</blockquote>
<p>What this means is that members of the DAO purchase (or otherwise receive) tokens that allow them to vote on decisions made by the organisation. Once the requisite number of votes is received, a smart contract (a small program) then takes the action that the vote is on. It becomes more complicated than that, but that’ll do for the sake of simplicity.</p>
<p>I sort of like the concept of DAOs? The idea of an organisation which has a pot of money provided by its investors or commercial activity which can take actions based on votes? That sounds good! It’s somewhat like a board within any company or similarly structured organisation but with additional anonymity and less regulation.</p>
<p>PleasrDAO describes themselves as a “<a href="https://www.youtube.com/watch?v=sC-35uHaMG8">worldwide crypto collective</a>”. Founded in early 2021, the group initially came together to purchase a piece of digital art by <a href="https://fortune.com/2021/08/06/nft-art-pplpleasr-fortune-cover-ethereum-defihow-crypto-changed-my-life/">pplpleasr</a>, an artist known for her work as one of the early creators of artwork as a Non-Fungible Token (NFT). Since then, they’ve expanded their mission to acquire culturally significant items, with a focus on tokenizing these assets and “<em>sharing ownership with a broader community</em>”.</p>
<h3 id="nfts">NFTs</h3>
<p>NFTs, and by that I mean specifically digital artwork that is available as an NFT, have always struck me as an intriguing but ultimately ridiculous part of the early 2020s wave of crypto-mania. JPGs that could be freely downloaded with a right-click, save as, being ascribed often huge values just to be able to say you ‘own’ that piece of art? I’m not a luddite but… no.</p>
<p>The attempts to compare ownership of digital artworks with a physical piece of art never made any sense – if you own a Monet, you own the version of the painting that Monet actually touched and painted. It is a genuine and unique one-off. A digital NFT is, by its very nature, not a one-off, and not even the same file that an artist originally pushed pixels around in. Let alone the fact that anyone can make a precise, exact copy of the piece of art whenever they want.</p>
<p>The concept of shared ownership of a physical piece of art using NFTs though? That sounds pretty cool! Sure, it’s long been possible to <a href="https://www.masterworks.com/">invest in an art-based fund</a> but not with the ease with which I can just buy an NFT. I can see a use-case here!</p>
<h3 id="the-great-wu-tang-nft-experiment">The Great Wu-Tang NFT Experiment</h3>
<p>So, back to The Album. PleasrDAO buy the album and sit on it for a couple of years whilst they work out a deal with Wu Tang Clan and figure out how to get a return on their investment; they’ve spent $4m on this album, but aren’t allowed to commercially exploit it for another 80 years.</p>
<p>Finally, this year they figure out a deal; they’ll sell “copies” of the album for $1 a piece. Each purchase will come with a 5-minute sampler of the album, and each purchase will accelerate the point at which the whole album can be released by 88 seconds. Presumably, once the release date is reached, the NFT owners will receive a copy of the album, although that isn’t actually explicitly stated.</p>
<p>What the NFT doesn’t give the purchaser, however, is a share of The Album itself – PleasrDAO still owns that outright, so if they decide to sell it tomorrow the NFT owners won’t get anything back.</p>
<p>So are people buying the NFT? Yes! The <a href="https://www.thealbum.com/leaderboard">website</a> shows that some 300,000 copies have been purchased and the release date has been accelerated by just less than a year to 2102.</p>
<p>We can assume that some people are buying the NFTs just to own the NFT (?) but the main drive is to bring the release date forward. For fans like me, the allure of the entire thing is not the intrinsic value of the album, it’s in hearing music that until now I’ve been excluded from hearing. To sate this desire, PleasrDAO have been hosting “listening parties” in Australia, with a special 30-minute mix <a href="https://en.wikipedia.org/wiki/Wu-Tang:_Shaolin_Style">played on a PlayStation 1</a>.</p>
<p>And herein lies the rub. The intrigue and excitement around the album, and therefore the value, is in the music itself rather than the objet d’art. As soon as the full album is released, the value of the original artefact, the disk and the pretty box, drops to (near) zero.</p>
<h3 id="the-shkreli-factor">The Shkreli Factor</h3>
<p>Which brings us back to Martin Shkreli. Shkreli poses an almost existential risk to PleasrDAO and their plans. It was inevitable that, given that the Album is a CD containing digital files, Shkreli made a copy.</p>
<p>On June 9th, Shkreli, in an unsurprising display of being himself, <a href="https://x.com/MartinShkreli/status/1799871344944849333/">hosted a Twitter/X Space</a> and streamed an hour of the album, prompting a lawsuit from <a href="https://pitchfork.com/news/martin-shkreli-sued-for-copying-and-playing-wu-tang-clan-album-without-permission/">PleasrDAO in response</a>. A judge then <a href="https://www.reedsmith.com/en/news/2024/06/pharma-bro-martin-shkreli-ordered-to-hand-over-all-copies-of-a-wutang-album">ordered him to hand over all copies of the album that he had</a>. For now, let’s assume that he did this, and more importantly that he deleted those copies<sup><a href="#user-content-fn-2" id="user-content-fnref-2" data-footnote-ref="" aria-describedby="footnote-label">2</a></sup>.</p>
<p>This episode highlights the inherent fragility of digital artworks as assets. While proponents of NFTs for digital artworks argue that “ownership” of the digital asset justifies the price, the reality is that the ease with which they can be copied brings their value down to zero once they are widely available. It is this risk that PleasrDAO are countering – they have a $4m asset that can have its value destroyed immediately as a result of the mercurial temper of a previous owner, despite him not having access to the original item for over 5 years.</p>
<p>The legal implications of Shkreli’s actions are fascinating. While he no longer owns the physical album, he retained a copy of its contents. This raises questions about the nature of ownership in the digital age. Can you truly “own” digital content if you can’t control its distribution? The judge’s order to hand over all copies suggests a legal system grappling with these new realities.</p>
<h3 id="the-saga-continues">The Saga Continues</h3>
<p>There is something relatively neat about this entire thing as an analogy for how I feel about Crypto in general. We have something that does, definitely have <em>some</em> value. It has then likely had its value greatly inflated by attaching it to NFTs and the Blockchain. And then, because of the absurdity of its history, the whole construct becomes insanely fragile – a bubble that can be burst at any moment, bringing the inflated value back down to near zero.</p>
<p>The entire value proposition rests on the assumption that everyone will play nice and follow the rules. In the world of crypto and finance, that’s a bet I wouldn’t take with someone else’s money, let alone $4 million of my own, especially when someone who is definitionally known for “not playing nice” is involved.</p>
<p>It feels inevitable that this all ends with the Album appearing online one Sunday afternoon, with dozens of people like me (but not me, you understand), returning to the simpler way of life of gleefully waiting for a torrent to finish downloading so that we can listen for the first time.</p>
<p>In the end, what have we learned? That rarity can create value, but also that humans have an infinite capacity to create artificial scarcity. That blockchain technology can be used to tokenize anything, even the right to maybe hear an album someday. And that no matter how much you hate on Martin Shkreli, it’s sometimes nice to have a proper villain in a story.</p>
<p>As for Wu-Tang Clan, I can’t help but wonder if this is what they had in mind when they set out to make a statement about the value of music. Did they anticipate that their album would become a case study in crypto economics and digital ownership? Probably not. But then again, Wu-Tang Clan ain’t nothin’ to fuck with, and neither, apparently, is the intersection of music and financial innovation.</p>
<p>In the meantime, I’ll be here, listening to “36 Chambers” on Spotify like a normie, wondering where it all went so beautifully, ridiculously wrong.</p>
<section data-footnotes="" class="footnotes"><h2 class="sr-only" id="footnote-label">Footnotes</h2>
<ol>
<li id="user-content-fn-1">
<p>It is actually a BVI-registered company <a href="#user-content-fnref-1" data-footnote-backref="" class="data-footnote-backref" aria-label="Back to content">↩</a></p>
</li>
<li id="user-content-fn-2">
<p>He did not, because of course he didn’t. Probably. <a href="#user-content-fnref-2" data-footnote-backref="" class="data-footnote-backref" aria-label="Back to content">↩</a></p>
</li>
</ol>
</section>
<p><em>Originally published at <a href="https://www.technicalchops.com/articles/wu-tang-forever">technicalchops.com</a>.</em></p>]]></content:encoded>
      <pubDate>Tue, 23 Jul 2024 00:00:00 GMT</pubDate>
      <author>simon@minton.co (Simon Minton)</author>
      <category>Crypto</category>
      <category>News</category>
      <category>Music</category>
    </item>
    <item>
      <title>AI Goes to Hollywood</title>
      <link>https://www.technicalchops.com/articles/ai-goes-to-hollywood</link>
      <guid isPermaLink="true">https://www.technicalchops.com/articles/ai-goes-to-hollywood</guid>
      <description>What could the arrival of AI mean for Hollywood?</description>
      <content:encoded><![CDATA[<p>New technology, of all kinds, is notoriously difficult to “put back in the box” once it’s out. Early-stage technologies tend to expand rapidly, often outpacing society’s ability to adapt. As a result, initial responses to these rapid advancements are typically reactive and focused on potential negative impacts.</p>
<p>The 2023 Hollywood strikes exemplified this phenomenon. While the <a href="https://stratechery.com/2023/hollywood-on-strike/">labour dispute</a> was rooted in the effects of streaming on the industry’s economic model, the perceived threat of Artificial Intelligence (AI) to filmmaking became an increasingly prominent issue.</p>
<p>But were the screenwriters and actors right to be concerned? The answer, as is often the case with emerging technologies, is complex. AI will undoubtedly change Hollywood, just as it will transform all industries. Let’s take a deep dive into the film industry to explore the potential effects of AI.</p>
<h2 id="screenwriting">Screenwriting</h2>
<h3 id="from-campfires-to-computers-the-evolution-of-storytelling">From Campfires to Computers: The Evolution of Storytelling</h3>
<p>To fully appreciate the role AI might play in the future of filmmaking, we need to step back and consider the broader context of storytelling itself. Storytelling is perhaps humanity’s oldest art form, a fundamental aspect of our species that predates written language, agriculture, and even permanent settlements.</p>
<p>The evolution of storytelling mirrors the march of technological progress:</p>
<ol>
<li>Oral tradition: Stories were fluid, adaptive, passed down through generations by word of mouth. Each telling was a unique performance, shaped by the teller and the audience.</li>
<li>Writing (c. 3200 BCE): For the first time, stories could be fixed, preserved exactly as their authors intended. This permanence allowed for more complex narratives, but limited access to the literate elite.</li>
<li>Printing press (15th century): Books could be mass-produced, making them affordable and accessible to a much wider audience. This laid the groundwork for the novel.</li>
<li>20th century media: Radio, television, and cinema brought new dimensions to storytelling, combining visual and auditory elements.</li>
<li>Internet age: New forms like interactive narratives and transmedia storytelling emerged. Content creation was democratised to an unprecedented degree.</li>
</ol>
<p>Each of these technological advancements expanded the reach and accessibility of storytelling. They provided new tools for creators and new experiences for audiences. But crucially, they didn’t replace human creativity – they enhanced it.</p>
<p>AI represents the next step in this evolution. But unlike previous technological advancements which primarily affected distribution and consumption, AI has the potential to both change distribution and impact the creation process equally. This is what makes it both exciting and terrifying to many in the industry.</p>
<h3 id="the-limit-of-ais-imagination">The Limit of AI’s Imagination</h3>
<p>AI’s ability to assist in creative processes is undeniable, yet it faces inherent limitations in generating truly original content. At their core, AI models—be they Large Language Models or image generators—are sophisticated pattern recognition systems, bound by the confines of their training data.</p>
<p>Consider an AI trained exclusively on black-and-white films from the early 20th century. This AI, despite its processing power, couldn’t conceive of colour cinema, let alone today’s immersive 3D experiences. This isn’t a technical shortcoming, but a conceptual one—the AI can only operate within the boundaries of its known world; the data that it has been trained on.</p>
<p>This limitation becomes critical when we examine AI’s role in creative fields. While AI can produce seemingly novel content, this novelty is merely a recombination or extrapolation of existing data. The truly groundbreaking ideas—those that redefine genres and push storytelling boundaries—remain out of AI’s reach, a direct consequence of its fundamental architecture.</p>
<p>As a result, the ability to create something original, films that transcend existing paradigms and construct entire cinematic universes from scratch, remains a uniquely human ability.</p>
<h3 id="avatar-pushing-the-boundaries-of-human-creativity">Avatar: Pushing the Boundaries of Human Creativity</h3>
<p>James Cameron’s ‘Avatar’ is a testament to the power of that originality. Released in 2009, it became the highest-grossing film of all time, a feat nearly matched by its 2022 sequel.</p>
<p>What makes Avatar’s success particularly striking is that it wasn’t an adaptation, remake, or comic-book franchise instalment; it was an entirely original film. Yes, critics pointed out echoes of familiar narratives like ‘Pocahontas’ or ‘Dances with Wolves’, but this missed the forest for the trees. The true innovation of Avatar lies not in its plot, but in its execution – the immersive world-building, the groundbreaking visual effects, and the overall cinematic experience that Cameron crafted.</p>
<p>Cameron didn’t just sit down and write a script. He spent years developing the world of Pandora, drawing inspiration from Earth’s deep-sea ecosystems, which he <a href="https://www.youtube.com/watch?v=bLU2tXj39K4">explored in his documentary work</a>. He collaborated with <a href="https://www.salon.com/2023/01/11/navi-language-avatar-linguistics-paul-frommer-interview/">linguists to create the Na’vi language</a>, with artists to design the flora and fauna of Pandora, and with engineers to develop the technology needed to bring his vision to life. He pushed the boundaries of what was <a href="https://screencraft.org/blog/the-insane-film-technology-behind-avatar-the-way-of-water/">technologically possible in filmmaking</a>, pioneering new motion capture techniques and 3D cinematography.</p>
<p>This level of holistic creativity – combining lived experience with storytelling and technological innovation – is something that current AI, bound by its training data, simply cannot replicate.</p>
<h3 id="human-connection">Human Connection</h3>
<p>That ability is not limited to film; the intricacies of the best long-form television series demonstrate that the best storytelling is predicated on building long-term human and emotional connections with audiences. Shows like ‘<em>Breaking Bad</em>’ and ‘<em>The Wire</em>’ weave complex narratives that span dozens of hours, balancing episodic stories with overarching season arcs and series-long character developments.</p>
<p>To replicate just this style of storytelling, an AI would not only require an enormous context window to manage entire season and series storylines – a feat that’s far beyond current capabilities – it would need to understand and replicate the subtle art of pacing, the careful revelation of information, and the delicate balance of multiple character arcs that make these series so compelling.</p>
<p>AI models, despite their complexity, lack the deep understanding of the human condition required to create these types of work. On the surface, ‘The Wire’ is a crime drama about the Baltimore drug scene. But over five seasons, it evolves into a profound exploration of institutional dysfunction, touching on education, politics, media, and the death of the American working class. The show’s creator, David Simon, drew on his experience as a crime reporter to create a narrative that was both entertaining and deeply insightful about American urban life.</p>
<p>Storytelling in this manner requires not just creativity, but also real-world knowledge, emotional intelligence, and a nuanced understanding of human society – attributes that AI, which can only work with the tangible data it’s been trained on, struggles to replicate.</p>
<p>Focusing on the best of human filmmaking and comparing it to the output of AIs is possibly a little disingenuous. These films and TV series don’t represent the mean, and certainly don’t represent the entire spectrum. By their very nature, they are the ‘standouts’.</p>
<p>There is a significant market for what might be termed ‘filler’ or ‘blockbuster bargain bin’ content; the lower-end, Hallmark channel comedies and rom-coms that follow well-established formulas. It is here that AI will likely have the largest impact on screenwriting. The ability for AI models to produce scripts that <a href="https://chatgpt.com/share/006f7d28-a0f3-4f7e-82eb-833e96e3df84">follow a well-trodden path</a> already exists and AI will likely lead to a glut of these straight-to-streaming action movies or holiday rom-coms that follow well-established formulas.</p>
<h2 id="performance">Performance</h2>
<p>This turns our attention to the performance aspect of filmmaking – the work of actors and directors in bringing a script to life. It is here that AI’s role becomes both more prominent and more contentious.</p>
<h3 id="quality-acting">Quality Acting</h3>
<p>The idea of AI or Computer-Generated Imagery (CGI) replacing human actors is not new territory.</p>
<p>Indeed, we’ve already seen impressive use of digital de-aging technology in films like ‘The Irishman’ and the creation of digital versions of deceased actors in ‘Rogue One’. However, it’s crucial to understand that these digital creations are ultimately based on human performances. In ‘Rogue One’, the digital recreation of Peter Cushing’s Grand Moff Tarkin was <a href="https://slate.com/culture/2017/02/guy-henry-becomes-peter-cushing-s-tarkin-in-this-rogue-one-special-effects-video-from-industrial-light-magic-video.html">based on a motion-capture performance by actor Guy Henry</a>. Computers didn’t create the performance – they merely helped translate a human performance into a digital model.</p>
<p>AI could dramatically enhance these capabilities, making it easier and more cost-effective to alter actors’ appearances or even create fully digital characters, but while it can replicate, manipulate, and, most recently, completely animate, it will continue to struggle with the kind of spontaneous creativity and emotional nuance that defines great acting. The chemistry of great casting matchups–Bryan Cranston and Aaron Paul, Brad Pitt and Angelina Jolie, or Simon Pegg and Nick Frost–is hard enough for humans to emulate, let alone for an AI.</p>
<p>We’re focusing on high-end cinema though; the best performances from the best actors. As with screenwriting, the harder effects of AI are most likely to be felt in the mid to lower end of the profession. With AIs able to do a passable job on generating performances at relatively low cost, many of these types of studios will likely turn to AI for a significant portion of their performance work, even if it is not the performances of the main characters.</p>
<p>The main cast are not the only actors to consider though; extras-heavy scenes, already often created in post-production with VFX, such as street or crowd scenes which would once have required a location shoot and dozens of extras and props, could now be generated in a snap.</p>
<p>With the ease of production of these types of films, we may see lower-end studios opting to put out tens of films a year produced in this way, or even see new types of hyper-personalised films emerge: A single base film could, via AI, become 1,000 different variations to appeal to niche markets. One single script and ‘shoot’ with human actors could result in many different films which appeal to various niches, with the base performances modified to feature characters with different hair colours, environments or even ethnicities.</p>
<h3 id="voice">Voice</h3>
<p>One area where AI might have a more transformative impact is in voice acting and dubbing. AI voice synthesis is advancing rapidly, and we might soon see AI systems capable of translating and dubbing films into multiple languages with the original actor’s voice while syncing the lip movements of the actors. This could make foreign language films more accessible to global audiences and potentially transform the economics of international film distribution.</p>
<p>Yet even here, the human element remains crucial. Voice acting isn’t just about producing the right sounds – it’s about bringing characters to life, conveying emotion and subtext through vocal performance. While AI might be able to handle more straightforward dubbing tasks, high-quality voice work is likely to remain the domain of human actors for the foreseeable future.</p>
<h3 id="solo-spielbergs">Solo Spielbergs</h3>
<p>One of the most exciting prospects of AI in filmmaking is the potential democratisation of high-quality content creation. AI tools could empower solo creators or small teams to produce today’s-Hollywood-calibre films from their bedrooms, dramatically lowering the barriers to entry for aspiring filmmakers. Very soon, individuals with creative vision but limited resources could compete with established studios in terms of production quality.</p>
<p>Imagine a world where a teenager with a compelling story could use AI tools to help them generate a screenplay, create stunning visual effects, and even produce realistic voice acting for multiple characters. MKBHD talks about needing to be an octopus as a solo creator - AI will give those creators access to robotic arms allowing them to create work that would otherwise be impossible by themselves. This level of democratisation could lead to an explosion of diverse voices and stories in the film industry.</p>
<h2 id="post-production">Post-Production</h2>
<p>Behind the scenes, the story becomes more complicated. While the past decades have seen huge growth in the world of VFX, it is apparent that AI will soon be an important tool within the arsenal of art directors.</p>
<p>Whilst current AI video production has issues with consistency, morphing and physics, the field is moving extremely quickly.</p>
<p>At the very least, it is now at a point where it can convincingly produce ‘B-roll’ footage which can then be edited into <a href="https://www.youtube.com/watch?v=KFzXwBZgB88">something that works</a>. This B-roll or background footage may have been shot by the ‘second unit’ or created in VFX, which will have a labour impact on those teams. How quickly it gets to the point where the more fine-grained control that is required for professional productions is available, however, is difficult to say.</p>
<p>As AI tools become more prevalent in filmmaking, we’re likely to see a significant evolution in what constitutes spectacle on screen. When every production has access to stunning visual effects and the ability to create fantastical worlds with relative ease, the bar for what impresses audiences will inevitably be raised.</p>
<p>The audience should be the beneficiary, with a potential arms race of visual innovation, with big-budget productions pushing the boundaries of what’s visually possible to differentiate themselves or the emergence of new forms of visual storytelling that take full advantage of AI’s capabilities.</p>
<p>With more creative freedom for filmmakers, as AI tools reduce the cost and complexity of realising their visions on screen, we may see a new wave of visually ambitious independent films, as creators who previously couldn’t afford extensive visual effects gain access to powerful AI tools.</p>
<p>On the other hand, as audiences become accustomed to high levels of visual spectacle across all types of content, we might see a renewed emphasis on storytelling and high-quality, real acting. When every film can offer stunning visuals, the ones that stand out will be those that offer something unique in terms of narrative, character, or thematic depth. Just as the rise of processed foods led to a renewed appreciation for artisanal, handcrafted products, the proliferation of AI in storytelling could lead to a counter-movement celebrating distinctly human creativity.</p>
<p>This could manifest in various ways; a continued resurgence of practical effects in film, with audiences appreciating the tangibility and imperfections of real-world stunts and sets, similar to the approach of both Christopher Nolan or Tom Cruise; a renewed interest in single-take shots or long, unbroken scenes that showcase actors’ raw performances and a renaissance in documentary filmmaking, with its focus on real human stories and experiences.</p>
<h2 id="distribution-and-consumption">Distribution and Consumption</h2>
<p>All of these shifts in the creative process could also necessitate entirely new distribution models.</p>
<p>With solo creators able to publish full films, we might see the emergence of aggregator platforms that combine elements of Netflix, YouTube, and social media, allowing creators to publish their works directly to viewers and leveraging both user feedback and AI-driven recommendation engines to match content with interested audiences.</p>
<p>The democratisation of film creation also presents challenges. As the barriers to entry lower, we’re likely to see a flood of content. This could make it even harder for truly original and compelling stories to break through the noise. The challenge of discovery – already a significant issue in the streaming era – could become even more pronounced. Unlike TikTok where the investment of time a creator asks for is 30 seconds to a minute, films require a greater willingness from the audience to engage with an unknown quantity.</p>
<p>Moreover, this democratisation could significantly impact the economics of the film industry. The traditional studio system, with its high barriers to entry and control over distribution, could face unprecedented challenges. We might see a shift towards a more decentralised model, with a long tail of independent creators alongside the big studios.</p>
<p>New ways of consumption that have deeper integrations with AIs that are, via smartwatches and other data sources, able to understand a viewer’s emotional state and environment, might allow platforms to either offer content that accords with that, or even to specifically generate content that works for you.</p>
<p>Such a platform could use AI to analyse viewer preferences and behaviours at a granular level to suggest the perfect content for each viewing session. This could create new opportunities for niche content to find its audience.</p>
<p>More interactive content, whether directly, in the same vein as Netflix’s Black Mirror:  <a href="https://en.wikipedia.org/wiki/Black_Mirror:_Bandersnatch">Bandersnatch</a>, or indirectly, with a film adjusting its pacing, tone, or even plot points based on real-time analysis of viewer engagement, could allow users to enjoy a new type of media that sits somewhere between a traditional film or a game.</p>
<p>We could also see the emergence of more personalised narratives, where AI tailors the story to individual viewers’ preferences, even going so far as having films that are adaptable to the individual viewer on the fly; a hero that looks like you, in the world that you know.</p>
<p>I doubt that the communal cinema experience is going anywhere anytime soon though and expect that there will be an increase in more immersive, IMAX-style, blockbuster experiences which aim to offer a highly differentiated product to audiences.</p>
<h2 id="the-economics-of-ai-in-hollywood">The Economics of AI in Hollywood</h2>
<p>The integration of AI into filmmaking will inevitably reshape the industry’s economics. This shift is likely to be as significant as the rise of streaming platforms, if not more so.</p>
<p>One of the most immediate impacts will be on production costs. AI has the potential to streamline nearly all aspects of production, especially in post-production.</p>
<p>The accessibility of complex visuals to all teams could also be a boon to mid-tier studios though. This could make visually ambitious films more feasible at lower budget levels, potentially revitalising the mid-budget sci-fi and fantasy genres.</p>
<p>The Hollywood strikes weren’t wrong; there will be an effect on the labour force, from scriptwriting to visual effects and everyone in between. It’s not all bad news though; while some roles will be displaced by AI, new positions focused on AI implementation and oversight will emerge. We might see the rise of ‘AI wranglers’ on film sets, experts who can interface between traditional filmmaking roles and AI systems.</p>
<p>The potential for AI-driven personalised content could create new revenue streams but might also fragment audiences. If viewers can get hyper-personalised content tailored to their specific tastes, will they still be willing to pay premium prices for big-budget, one-size-fits-all blockbusters?</p>
<p>As AI becomes capable of generating content based on existing intellectual property, we might see changes in how IP is valued and protected. The ability of AI models specifically trained on a franchise to create new stories in established universes could make these already valuable franchises even more lucrative. The flipside is that, as we have seen with Disney’s <a href="https://screenrant.com/star-wars-disney-criticism-explained-matthew-belloni/">mixed success in their huge growth of the Star Wars universe</a>, fans sometimes actually prefer it when they don’t feel that their favourite stories are turned into derivative and low-effort output.</p>
<p>Moreover, as AI-generated content becomes more prevalent, we might see the value of original, human-created stories increase. Just as the rise of streaming led to bidding wars for top showrunners and content creators, we might see studios placing a premium on distinctly human creativity as a way to stand out in an AI-saturated market.</p>
<p>The international market, already a crucial consideration for major studios, could become even more significant. AI’s potential to easily dub and localise content could make international distribution more seamless, potentially leading to more globally oriented production strategies.</p>
<p>AI will present challenges to Hollywood and the filmmaking industry, particularly in terms of job displacement and market disruption, but it’s also going to create new opportunities. The studios and creators who thrive in this new landscape will be those who can effectively harness AI’s capabilities while maintaining the human creativity and vision that ultimately drives the art of cinema.</p>
<p><em>Originally published at <a href="https://www.technicalchops.com/articles/ai-goes-to-hollywood">technicalchops.com</a>.</em></p>]]></content:encoded>
      <pubDate>Fri, 28 Jun 2024 00:00:00 GMT</pubDate>
      <author>simon@minton.co (Simon Minton)</author>
      <category>AI</category>
      <category>Media</category>
      <category>Long Term Effects</category>
      <category>LLMs</category>
    </item>
    <item>
      <title>Apple Intelligence: Context Matters</title>
      <link>https://www.technicalchops.com/articles/apple-intelligence-context-matters</link>
      <guid isPermaLink="true">https://www.technicalchops.com/articles/apple-intelligence-context-matters</guid>
      <description>Apple&apos;s Artificial Intelligence is finally here</description>
      <content:encoded><![CDATA[<p>Today, Apple announced the arrival of Apple Intelligence (AI) – the long-awaited integration of Artificial Intelligence (AI) into their main product lineup of iOS devices. Accompanied by the faint whisper of 100 more AI app startups taking their last breath, the announcement was a declaration of intent; the next version of iOS will be deeply infused with AI. In my <a href="https://www.technicalchops.com/articles/how-hard-is-being-an-assistant-anyway/">previous exploration</a>, I posited that the utility of an AI assistant depends on its ability to have personal context and deep knowledge about my life and preferences. Without that understanding, AI assistants are of relatively little use.</p>
<p>The most promising element of AI for me, therefore, is the depth of integration and access promised for the AI. Apple’s approach is unique – by allowing what is, in effect, two AIs to work side by side, with a private, local-only AI having access to data on your phone and a secondary (OpenAI-powered) AI acting as a higher-powered interaction layer, they have created a proposition that could actually achieve what I had hoped for in <a href="https://www.technicalchops.com/articles/apple-and-ai">Apple and AI</a>.</p>
<p>As a result of the promised deep integration, Apple Intelligence will be able to access a level of personal context that apps just can’t match. And that ability to access personal context without requiring direct user input <strong>really</strong> matters. To understand why it matters so much, we need to consider how AIs obtain, retain, and process information.</p>
<h2 id="traditional-computing-vs-ai">Traditional Computing vs AI</h2>
<p>In computing, there have traditionally been two main ways that information is stored: on a hard drive and in RAM (Random Access Memory). In the mental model that we have of our own memory, a hard drive stores long-term memories—knowledge that you know, a visual scene of your 12th birthday party, your date of birth—and RAM stores your working or short-term memory—what you need to get from the shops that evening, or what you <a href="https://en.wikipedia.org/wiki/Doorway_effect">went to the kitchen to do</a>.</p>
<p>AI models work quite differently from a traditional computer. Yes, the models are stored on a hard drive, and most of the work is done “in memory,” but the model doesn’t specifically draw on files to access information. Instead, models effectively have two separate elements: the statically trained model itself and the ‘context window’.</p>
<h2 id="long-term-memory-training-data">Long-Term Memory: Training Data</h2>
<p>When we ‘train’ a model, we provide it with what is described as a ‘pile’ of information. This pile, or large dataset, is what the model learns patterns and information from. Effective and extensive training gives the model behind ChatGPT (other models are available) the ability to accurately answer when World War II occurred, for example, based not on a file that tells it the correct dates but from the patterns it learned from within the training data.</p>
<p>Whilst it is being trained, the model learns the structure of language from the corpus, which is what gives it the ability to respond correctly in different languages. There are <a href="https://huggingface.co/blog/rlhf">other processes</a> that enhance the model’s performance, but this provides a basic understanding of how information is embedded into a model.</p>
<p>The way a model retrieves information and constructs sentences for its answers are intertwined processes. It generates an answer that is probabilistically correct for the given query, based on the patterns and knowledge it has learned from the training data.</p>
<p>This approach works quite well. We now have models that have been trained on extremely large datasets that appear to know a lot of information and have a great command of many languages—these are the models at the forefront of generalised AI. The training process for a model is relatively slow, though, and (at the moment) that training can’t be kept continuously updated, so even the most recently released models have a relatively out-of-date knowledge cut-off date. The information stored in the model is also impersonal in nature – it doesn’t specifically know about you and your life, and when you do directly provide it with information, you are not adding to the model itself. Instead, you are working within what we consider to be the AI’s short-term memory.</p>
<h2 id="short-term-memory-context-windows">Short-Term Memory: Context Windows</h2>
<p>This short-term memory is the ‘context window’. The term refers to the amount of new information that an LLM can understand at once. Whilst it appears that we are in a back-and-forth conversation with a model when we’re using these tools, in fact, we are starting afresh with every request each time we send a new message to ChatGPT. What is happening behind the scenes is that the entire conversation to that point is being sent to the model each and every single time, to provide it with the historical context of the conversation. The model itself remains unchanged by your inputs; it is just provided with the greater context afforded by having the previous chat.</p>
<p>Context windows are called windows for a reason, though; they have bounds, and beyond a certain point, the historic conversation will be cut off to fit within the window. For GPT-4, that point is at 128,000 tokens, or about 96,000 words. For Gemini 1.5 Pro, it’s as much as 2 million tokens or around 1.5 million words. This sounds like a lot, but in the context of your phone’s data, it barely scratches the surface.</p>
<p>There is another approach though; providing the AI with the ability to pull in data from other authoritative data sources that are relevant in the moment - we call this Retrieval-Augmented Generation (RAG). By pulling in data based on relevance to a given query from other sources, the limitations of both a lack of trained knowledge and a small context window can be overcome. And it is this approach that I had hoped that Apple would take from when I first discussed this earlier this year.</p>
<h2 id="the-apple-difference">The Apple Difference</h2>
<p>I’ve previously expressed scepticism regarding AI assistants, partly because current versions of LLMs lack the personal contextual awareness I would want from an assistant. Even where an assistant is able to persist ‘important’ information, there are so many pieces of personal context that an AI without access to your entire life simply can’t have. That context is stored in messages, calendars, emails or even just the places that we’ve been to or currently are. To achieve the level of context required to be truly useful without needing to be constantly updated by the user requires something that few devices can actually have; a persistent and all-seeing presence in your life. It turns out that the best way to have that persistent presence is to be your most persistently present personal device; your phone.</p>
<p>What makes AI exciting is its ability to integrate personal context into the model without needing to directly provide the information. By drawing from its access to your phone’s personal data, the AI will be able to understand more about your life than any other assistant would be able to. This is as a direct result of the positioning of the AI within the stack – it sits at the OS level and can therefore do things that no standalone app could ever do.</p>
<p>Apple has also spent no small amount of time and effort focussing on not only branding itself as a privacy-conscious platform, but also ensuring that the choices that it makes within its platforms are genuinely driven by privacy. They have built a level of user trust that is unmatched. That Apple is able to run their AI on-device matters – personal information isn’t shared to the cloud and stays safely, and securely, on-device. It really is only Apple that can provide this due to their unique approach to hardware and software.</p>
<p>If iOS 18 delivers on today’s promise, I’m excited. This is the first time that I’ve felt any company has truly understood how to approach AI in a way that is genuinely helpful and understands what an AI assistant needs to be more than just a toy. If you thought I was bullish on Apple prior to this event…</p>
<p><em>Originally published at <a href="https://www.technicalchops.com/articles/apple-intelligence-context-matters">technicalchops.com</a>.</em></p>]]></content:encoded>
      <pubDate>Mon, 10 Jun 2024 00:00:00 GMT</pubDate>
      <author>simon@minton.co (Simon Minton)</author>
      <category>AI</category>
      <category>Assistants</category>
      <category>Agents</category>
      <category>Mobile</category>
      <category>Long Term Effects</category>
      <category>LLMs</category>
      <category>Apple</category>
    </item>
    <item>
      <title>AI Inputs and Outputs in the Real World</title>
      <link>https://www.technicalchops.com/articles/ai-inputs-and-outputs-in-the-real-world</link>
      <guid isPermaLink="true">https://www.technicalchops.com/articles/ai-inputs-and-outputs-in-the-real-world</guid>
      <description>Creating objects in the real world with AI</description>
      <content:encoded><![CDATA[<p>On Tuesday, <a href="https://blog.google/inside-google/message-ceo/google-io-2024-keynote-sundar-pichai/#gemini-era">Sundar Pichai opened the Google I/O keynote</a>:</p>
<blockquote>
<p>”A year ago on the I/O stage we first shared our plans for Gemini: a frontier model built to be natively multimodal from the beginning, that could reason across text, images, video, code, and more. It marks a big step in turning any input into any output — an “I/O” for a new generation.”</p>
</blockquote>
<p>Multimodal AI has been an obsession of mine for years. When we launched <a href="https://www.technicalchops.com/articles/webinars-and-the-evolution-of-playbox/">Playbox</a>, the premise was clear: all content mediums should be interchangeable. The vision of converting a video to an article or vice versa was tantalisingly close, yet elusive in 2020. Fast forward to today, and it appears we were more prescient than overly optimistic.</p>
<p>OpenAI’s <a href="https://openai.com/index/spring-update/">event</a> (notably announced with very short notice to occur less than 24 hours before Google’s) on Monday showcased a new, faster model: GPT-4o. This appears to be the first truly multimodal model; it takes inputs of audio, video, text, or images directly, without an intermediary translation layer. Previous models required separate steps to handle different types of data, but GPT-4o integrates these capabilities natively.</p>
<p>The standout feature was the voice interaction. One of the challenges we faced with Playbox was the inadequacy of text-to-speech (TTS) technology at the time. Our custom models could imitate a person’s voice, but the result was often slightly robotic. The initial version of ChatGPT’s voice mode showed promise, but I <a href="https://www.technicalchops.com/articles/short-thoughts-on-chatting-with-ai/">wanted it to be more human</a>. The latest model, however, is almost disconcertingly human. Gone are the long, uninterruptible monologues after a brief wait. Now we have instantaneous responses, shorter answers, and the ability to cut off and redirect the AI simply by speaking. It is also, curiously, quite flirtatious. The voice interaction feels natural, with subtle meandering and emotional nuance, eerily similar to Scarlett Johansson’s voice in the movie <em>‘<a href="https://x.com/sama/status/1790075827666796666">her</a>’.</em><sup><a href="#user-content-fn-1" id="user-content-fnref-1" data-footnote-ref="" aria-describedby="footnote-label">1</a></sup></p>
<p>This model, and Google’s evolution of the Gemini model mean that AI can convert most digital inputs into human-readable digital outputs with remarkable consistency, albeit still with some errors. Whether it’s text, an image, a video, or sound, the AI can interpret and transform it into the desired format. That output however is by its nature creative and accommodates a degree of inaccuracy—success metrics for most outputs here are often relatively subjective.<sup><a href="#user-content-fn-2" id="user-content-fnref-2" data-footnote-ref="" aria-describedby="footnote-label">2</a></sup> This “creativity” of course is the result of probabilistic computing and is baked into the very nature of how AI works. Most of real life however is deterministic; a given set of inputs should result in a given output when an operation is applied.</p>
<h2 id="ai-and-3d">AI and 3D</h2>
<p>Hidden within the <a href="https://openai.com/index/hello-gpt-4o/">GPT-4o launch article</a> was an intriguing development: the model’s ability to produce 3D renderings.</p>
<p><img src="https://cdn.openai.com/hello-gpt-4o/3d-03.gif?w=640&#x26;q=90&#x26;fm=webp&#x26;fit=pad" alt="A spinning 3D OpenAI Logo"></p>
<p>A spinning 3D OpenAI Logo</p>
<p>The documentation mentions a 3D reconstruction based on six keyframes but lacks detail on the creation process. It suggests two possibilities:</p>
<ul>
<li>GPT-4o might include video capabilities to produce a consistent 3D-like video by interpolating between 2D frames.</li>
<li>Alternatively, GPT-4o might generate actual 3D files from images.</li>
</ul>
<p>While generating interpolated 2D frames is impressive, creating true 3D models would be revolutionary, not least because accurate 3D models require a level of determinism that we have not seen previously within large language models.</p>
<h2 id="bringing-ai-outputs-into-the-real-world">Bringing AI Outputs into the Real World</h2>
<p>If a model can take keyframes and build a 3D model from them, the implications for AI-generated outputs are immense. This advancement transitions us from digital constraints to generating tangible objects in the real world.</p>
<p>Over the past decade, 3D printing has become increasingly mainstream. From hobbyists’ spare bedrooms to Formula 1 labs, 3D printers enable rapid iteration and production of physical objects. Although industry and early adopters have embraced the technology, it has not yet become commonplace. This is primarily due to two factors: a) 3D modeling is complex and has a steep learning curve, and b) 3D printers are still not as user-friendly as the mass consumer market demands.</p>
<p>An AI model capable of accurately producing or reproducing physical objects from limited information could not only lower the barrier to creating 3D models for printing but eliminate it entirely. By combining this capability with more detailed input data—such as photos enhanced with mm-accurate LIDAR scans from high-end phones or stereoscopic views from multiple cameras—almost anyone could create 3D objects simply by taking a photo and sending it to the AI for reconstruction.</p>
<p>The major hurdle to overcome is the development of a 3D printer that is effortless to use. However, if creating 3D objects becomes easy for consumers, it is likely that printer manufacturers will rise to meet the demand, similar to how the rise of digital photography led to a surge in high-quality home photo printers.</p>
<p>As with any AI conversation, significant legal and ethical implications arise. If consumers can recreate objects simply by taking a photo and sending it to a printer, concerns about infringement will extend beyond artists, as seen with image generators like Midjourney and DALL-E. Companies will also worry, recognising that non-complex products could be easily replicated at home. Furthermore, questions will arise about the accountability of companies behind AI models that produce 3D objects, especially if an AI-designed device causes injury.</p>
<p>Whether GPT-4o can create true 3D objects or not, it feels like we are on the brink of a revolution. The convergence of multiple complementary technologies heralds an era where creating almost any output is becoming increasingly accessible. While we are not quite in the realm of Star Trek’s replicator, we are getting closer. The potential for AI to transform how we create and interact with both digital and physical worlds is enormous, and the implications—both exciting and challenging—will shape our future in profound ways.</p>
<section data-footnotes="" class="footnotes"><h2 class="sr-only" id="footnote-label">Footnotes</h2>
<ol>
<li id="user-content-fn-1">
<p>If you haven’t seen ‘her’ it’s definitely worth a watch given how closely reality appears to hewing to reality of the film. <a href="#user-content-fnref-1" data-footnote-backref="" class="data-footnote-backref" aria-label="Back to content">↩</a></p>
</li>
<li id="user-content-fn-2">
<p>Yes, AI is very good at generating code, which one could argue either objectively works or doesn’t but most often any significant code block that is generated by an AI needs a fairly high amount of human input to make sure it is actually fit for purpose… ask me how I know. <a href="#user-content-fnref-2" data-footnote-backref="" class="data-footnote-backref" aria-label="Back to content">↩</a></p>
</li>
</ol>
</section>
<p><em>Originally published at <a href="https://www.technicalchops.com/articles/ai-inputs-and-outputs-in-the-real-world">technicalchops.com</a>.</em></p>]]></content:encoded>
      <pubDate>Fri, 17 May 2024 00:00:00 GMT</pubDate>
      <author>simon@minton.co (Simon Minton)</author>
      <category>AI</category>
      <category>Long Term Effects</category>
      <category>LLMs</category>
      <category>3D</category>
      <category>Hardware</category>
    </item>
    <item>
      <title>Will AI kill sites like Booking.com? AI as a Personal Aggregator</title>
      <link>https://www.technicalchops.com/articles/ai-as-a-personal-aggregator</link>
      <guid isPermaLink="true">https://www.technicalchops.com/articles/ai-as-a-personal-aggregator</guid>
      <description>What will the arrival of personal AIs actually result in?</description>
      <content:encoded><![CDATA[<p>I talked a little bit about my thoughts on <a href="https://www.technicalchops.com/articles/how-hard-is-being-an-assistant-anyway/">AI Personal Assistants previously</a>. I find the idea very exciting! My general sense is that they are more complex to implement than people first imagine, but will be brilliant for relatively simple tasks that don’t involve a high degree of personal preference. What I didn’t explore in that article is the effect of their successful widespread adoption.</p>
<h2 id="ai-as-a-personal-assistant">AI as a Personal Assistant</h2>
<p>The interaction model suggested for these personal assistants is fairly straightforward: you say “<em>Get me a taxi there</em>” or “<em>Bring me a pizza here</em>” and any AI Assistant worth its salt will be able to translate your request into a series of necessary actions and negotiate with whichever third-party services are required to achieve the desired outcome.</p>
<p>If we ask an AI assistant to book a taxi, the AI itself will then interact with another service, whether via an API or (as the rabbit r1 does) directly interacting with websites and apps on your behalf, parsing the result and then feeding that back to you in the form of your conversation.</p>
<figure>
  <img src="https://www.technicalchops.com/images/lam.png" alt="Large Action Model">
  <figcaption>The rabbit r1 has a “Large Action Model” (LAM). The LAM translates your request into an action that it can perform on any given app, with the ability to learn how to perform actions on Apps that it hasn’t already been trained on.</figcaption>
</figure>
<p>For the rabbit r1, that means connecting your existing services to the platform. If you want your AI to be able to book a taxi then you will need to connect your Uber account first. The same with DoorDash for a pizza or Spotify for music.</p>
<p>That appears to be a perfectly sensible approach. We hand over control of our other accounts so that we can let our assistants use them on our behalf, as we might a human assistant. Now, instead of tapping on a screen, we’ll just speak or type and get exactly what we want from the connected apps.</p>
<p>Under the above model, we’re applying an existing and accepted paradigm to the new technology. In this case, we’re replicating what a human would do; open an app and get a result. The perceived benefit is in our own interaction – or lack thereof – with the end service.</p>
<p>From <a href="https://stratechery.com/2020/the-tiktok-war/">Stratechery</a>:</p>
<blockquote>
<p>When it comes to designing products, a pattern you see repeatedly is copying what came before, poorly, and only later creating something native to the medium.</p>
</blockquote>
<p>It feels like this is what is happening here. Connecting the AI to an app is just a copy of what came before. So, what is the native approach that is only possible as a result of the new technology?</p>
<p>Well, the huge benefit of an AI is not that it can simply replicate human functionality as a human assistant would, it’s that it can blend that with the more “traditional” functionality of computers that humans are not necessarily good at.</p>
<p>One of those things is doing multiple things, at the same time, very quickly.</p>
<h2 id="aggregators-and-the-status-quo">Aggregators and the Status Quo</h2>
<p>It won’t surprise you to learn that very few services on the internet are truly benevolent gifts to society. Wikipedia perhaps. <a href="http://Archive.org">Archive.org</a>, maybe.</p>
<p>Most websites are balancing their own motivation to make money with the quality of the service that they provide to users. Google is great, but the days of <a href="https://www.fastcompany.com/3060044/how-google-search-got-past-10-blue-links">10 blue links</a> are long gone – the results for any mildly competitive search term are headed with at least a screen of ads. Price comparison sites like Booking.com appear to exist to make customers’ lives easier but their revenue comes from the supply side, typically in the form of referral fees or ads.</p>
<p>Ben Thompson’s <a href="https://stratechery.com/2017/defining-aggregators/">Aggregation Theory</a> defines how these businesses work and the framework that they operate under. The internet is full of examples of companies that have, by securing both sides of a market, built incredibly successful businesses. One of his primary examples of an aggregator is Uber.</p>
<p>The initial technical idea behind Uber (<em>“a taxi at the tap of a button, with live tracking on a map and easy payment”</em>) was a good one! At launch, it was also relatively difficult to replicate. But not impossibly hard. Uber knew this; technology was not a moat on which it could rely. To build a real moat, they needed to invest in aggregating both sides of the market:</p>
<p><strong>Supply (Driver Network):</strong> Uber spent a lot of money rapidly building an extensive driver network to serve its app, ensuring there were always drivers available. They did this initially by paying very well, with high fees on journeys, bonuses and referral rewards.</p>
<p><strong>Demand (Passengers):</strong> Uber aggressively invested in marketing and branding to build its customer base, especially in building referral networks (at one point I had hundreds of pounds of credit on the app, just from these referrals). Uber’s prices when they launched were also <strong>ridiculously</strong> low. When Uber launched in London, a journey in an Uber Black Mercedes S-Class (the only Uber that was available) was cheaper than a Black Cab.</p>
<p>Uber <a href="https://www.bloomberg.com/news/articles/2018-03-06/uber-spent-10-7-billion-in-nine-years-does-it-have-enough-to-show-for-it">spent billions of dollars</a> executing all of these strategies and through its aggregation of the supply of taxis and passengers in a given locale, “won” many major cities. But that win came at a huge cost, subsidised by investors. Eventually, it had to normalise costs; driver pay went down and passenger costs went up. The moat that they’d built was wide and deep, but competitor services popped up and built their own networks by offering similar incentives as Uber had.</p>
<p>The key element for Uber though is that it still has the most users, who return to the app because it still has the most available taxi drivers, who in turn return to the app because it has a consistent supply of customers.</p>
<p>The actual service that Uber offers though is not highly differentiated. In any given major metropolitan area, there are at least 4 or 5 significant apps that provide the same service –  routing a Toyota Prius to the customer to take them where they want to go – at around the same price point. Ultimately their product is a commodity, with little differentiation from other competitive apps. They, to some extent, rely on customers’ (a) habitual use of their app and (b) lack of motivation to check every service at once for the best price/arrival time.</p>
<p>What happens though if we introduce an intermediary into the equation which <em>is</em> motivated to check every service for the best deal? Say, an AI which can do multiple things, at the same time, very quickly?</p>
<h2 id="ai-as-a-personal-aggregator">AI as a Personal Aggregator</h2>
<p>This is the big paradigm shift: introducing the Personal Aggregator. Instead of a human user using one platform to look for a taxi, why would AI not look at <em>all</em> of the platforms to find the best match?</p>
<p><img src="https://www.technicalchops.com/images/Multi_App.png" alt="Multi_App.png"></p>
<p>Of course, this doesn’t apply just to taxis. A sufficiently powerful AI will be able to rely not just on the results of booking.com, Amazon (not technically an aggregator) or DoorDash, but the <em>entire internet</em>.</p>
<p>The fundamental shift here is in the motivation of AI itself. An AI Assistant is likely part of the user’s personal infrastructure – not that of a third party motivated by the desire to make profit. Of course, there is an assumption here, but I think it’s a fair one. The early generalised intelligence that we’re seeing produced by the likes of OpenAI isn’t focussed on doing just one task well, it does everything, from history homework and language translation to coding and eventually price comparison across the internet.</p>
<p>With a motivation of always returning to the user the best possible deal for a given stay, product or other service – without concern for who the seller is - and having looked at <em>the entire market</em>, an AI assistant stands to be the <strong>ultimate</strong> aggregator.</p>
<p>As part of our personal infrastructure, the motivation of the AI will be to go wherever it needs to to get the right deal, using whatever blend of requirements that the user has provided it with along with the historical preferences that it has learnt about you. For our taxi example, one person’s personal aggregator may aim for the cheapest possible car at the cost of comfort and expediency. The same AI acting as a personal aggregator for someone else might prioritise comfort over cost and so on.</p>
<p>The fundamentals of how we intertact with a huge array of businesses could completely shift, not just because using the AI is easier, but also because it is the best way to secure the best deal. The widespread adoption of this type of interaction with an AI as personal aggregator could have massive consequences for all sorts of businesses, and not just the existing aggregators.</p>
<p>If we go beyond the creation of an increasingly competitive (but arguably more efficient) market, the intermediating effect of the AI will disrupt one of the core tenets of Aggregator Theory: that of the direct relationship with the end user. When that direct relationship is removed in favour of the ease of use and better user experience of the AI, it is not just the transaction that is being intermediated, nor the buying decision, it is also the pre-buying decision. What is the point in marketing a commodity product from Store X to buyers if they’re going to buy it via their AI?</p>
<p>It may feel that we’re a while off this reality, but as Bill Gates said <a href="https://www.duperrin.com/english/2018/11/08/we-always-overestimate-the-change-that-will-occur-in-the-next-two-years/#:~:text=Which%20reminds%20me%20of%20that,was%20overestimated%20in%20the%201990s">“We always overestimate the change that will occur in the next two years and underestimate the change that will occur in the next ten”</a>.</p>
<p>The effects on the markets that these commodity products exist within will be drastic, but there will be an increased incentive for differentiation to occur.</p>
<h2 id="embracing-the-new-paradigm">Embracing the New Paradigm</h2>
<p>The advent of AI as a personal aggregator heralds a transformative era in how we interact with and perceive the digital marketplace. This shift isn’t just about convenience or efficiency; it’s about a fundamental change in the power dynamics between service providers, aggregators, and consumers. AI won’t necessarily be great at telling us <em>what</em> to buy, but it will be at telling us <em>where</em> to.</p>
<p>For decades, aggregators have thrived by simplifying choice and monopolising user attention. They’ve dictated market terms, often at the expense of both service providers and consumers. However, as AI steps in as the ultimate intermediary, it will potentially democratise access to the market. By evaluating all available options to identify the best deal, AI will disrupt the traditional aggregator model, shifting the focus back to the quality and value of the service itself.</p>
<p>This change will compel service providers to innovate and differentiate. Yes, initially there may be a race to the bottom in terms of price but differentiation will need to occur instead as a result of focusing on enhancing quality and user experience.</p>
<p>Moreover, the rise of AI assistants as personal aggregators raises important questions about the future of marketing and customer relationships. As AI becomes the gatekeeper of consumer choices, traditional marketing strategies might lose their efficacy. Businesses will need to adapt by finding new ways to engage with both AI systems and their users, focusing more on the intrinsic value of their offerings rather than just visibility and brand recognition.</p>
<p>The emergence of AI as a personal aggregator is not just a technological advancement; it is potentially a catalyst for a more equitable and consumer-centric marketplace. It will challenge existing business models and forces a rethink of market strategies across various sectors. We’re not there quite yet, but we are standing on the brink of a paradigm shift and it is essential for businesses to understand and adapt to these changing dynamics. The future promises an era where choice is not just about availability, but about relevance and quality, in which consumers are hopefully empowered by AI like never before.</p>

<p><em>Originally published at <a href="https://www.technicalchops.com/articles/ai-as-a-personal-aggregator">technicalchops.com</a>.</em></p>]]></content:encoded>
      <pubDate>Sun, 18 Feb 2024 00:00:00 GMT</pubDate>
      <author>simon@minton.co (Simon Minton)</author>
      <category>AI</category>
      <category>Aggregators</category>
      <category>Long Term Effects</category>
      <category>LLMs</category>
    </item>
    <item>
      <title>Generative Video: Targeting You Soon</title>
      <link>https://www.technicalchops.com/articles/generative-video-targeting-you-soon</link>
      <guid isPermaLink="true">https://www.technicalchops.com/articles/generative-video-targeting-you-soon</guid>
      <description>A huge leap forward in generative video</description>
      <content:encoded><![CDATA[<p>On Thursday OpenAI released <a href="https://openai.com/sora">Sora</a>, their text-to-video model. It is remarkable, and I plan to write more about it and the future of video more generally in a forthcoming article.</p>
<p>Here’s an example of what it can do…</p>
<p>Prompt:</p>
<blockquote>
<p>A stylish woman walks down a Tokyo street filled with warm glowing neon and animated city signage. She wears a black leather jacket, a long red dress, and black boots, and carries a black purse. She wears sunglasses and red lipstick. She walks confidently and casually. The street is damp and reflective, creating a mirror effect of the colorful lights. Many pedestrians walk about.</p>
</blockquote>
<p>Output:</p>
<video controls>
  <source src="https://cdn.openai.com/sora/videos/tokyo-walk.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
<p>This video was generated completely by the AI model. No 3D rendering or other inputs. It is, simply put, <em>lightyears</em> ahead of what we’ve seen from the likes of Google’s <a href="https://lumiere-video.github.io/">Lumiere</a> and Runway’s <a href="https://research.runwayml.com/gen2">Gen-2</a> which <em>were</em> the state of the art.</p>
<p>The most impressive details aren’t on the main Sora page though, they’re hidden away in the <a href="https://openai.com/research/video-generation-models-as-world-simulators">technical report</a>. Not only can Sora generate entirely fresh video as shown above, but it can also transform existing videos <strong>significantly</strong>, based on a text prompt:</p>
<div class="flex flex-col md:flex-row">
<div class="w-full md:w-1/2">
<p><strong>Input:</strong></p>
<video controls>
  <source src="https://cdn.openai.com/tmp/s/edit/base.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
<div class="w-full md:w-1/2">
<p><strong>Output:</strong></p>
<video controls>
  <source src="https://cdn.openai.com/tmp/s/edit/1.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
<blockquote>
<p>change the setting to the 1920s with an old school car. make sure to keep the red color</p>
</blockquote>
</div>
</div>
<p>There are obviously huge implications to this for video entertainment, but one of the more fascinating aspects of this for me is in what it means for ad creative.</p>
<p>Yes, we might not like video ads but they are sadly here to stay. One of the reasons we don’t like them (beyond the interruption) is that they often feel so un-targeted and repetitive. That’s not entirely surprising given that the cost of making a 30-second commercial runs into the 10s of thousands of dollars let alone the actual time it takes to produce.</p>
<p>From the preview that we saw on Thursday though, these videos can be made in less than 10 minutes and at <em>significantly</em> reduced cost. Whilst the prospect of seeing hyper-targeted videos akin to last decade’s <a href="https://www.vox.com/the-goods/2018/9/25/17887796/facebook-ad-targeted-algorithm">oddly specific t-shirt ads</a> is not appealing, from a brand perspective being able to quickly and cheaply produce not only a series of ads targeted at each and every demographic but also alternatives to A/B test with is a marketer’s dream.</p>
<p>Film one video and then have AI produce modified versions for different ages, locales and cultural norms – a little dystopian? Probably. Inevitable? Absolutely.</p>
<p>Sure, we’re (maybe) unlikely to see this footage used in the Super Bowl, but as YouTube ads? Absolutely. And really, I’m not entirely averse to receiving YouTube ads that fit with my interests as well as Instagram ads tend to.</p>
<p>It’s not just preplanned creative that will benefit here either – social media teams seeking to produce responsive ads that match the calibre of <a href="https://twitter.com/Oreo/status/298246571718483968">Oreo’s famous 2013 tweet in response to a power cut at the Super Bowl</a> now have a tool that can produce high-production quality video in minutes.</p>
<p><em>Originally published at <a href="https://www.technicalchops.com/articles/generative-video-targeting-you-soon">technicalchops.com</a>.</em></p>]]></content:encoded>
      <pubDate>Fri, 16 Feb 2024 00:00:00 GMT</pubDate>
      <author>simon@minton.co (Simon Minton)</author>
      <category>AI</category>
      <category>Media</category>
      <category>Advertising</category>
      <category>Long Term Effects</category>
      <category>LLMs</category>
    </item>
    <item>
      <title>Apple and AI</title>
      <link>https://www.technicalchops.com/articles/apple-and-ai</link>
      <guid isPermaLink="true">https://www.technicalchops.com/articles/apple-and-ai</guid>
      <description>Apple is further ahead with AI than most think</description>
      <content:encoded><![CDATA[<p>Media speculation has been swirling around the idea that <a href="https://www.bloomberg.com/news/newsletters/2023-04-18/is-apple-falling-behind-in-the-ai-race-against-microsoft-google-amazon">Apple is lagging behind</a> in the AI race.</p>
<p>The headline advances in AI over the past couple of years have been generative technologies, and if you’re comparing chatbots then, yes, Siri looks like ancient technology. Apple’s approach to AI though has been to infuse it throughout iOS, with the goal of incrementally improving the user experience throughout without highlighting to users just how much of the heavy lifting the AI is actually doing.</p>
<p>This is consistent with the Steve Jobs playbook: <a href="https://www.youtube.com/watch?v=dI93BvrBxQ0">“You’ve got to start with customer experience and work backwards to the technology. You can’t start with the technology.”</a> Apple aren’t focused on making AI the centrepiece of their product launches and in fact <a href="https://arstechnica.com/information-technology/2023/06/at-apples-wwdc-keynote-ai-never-came-up-by-name-but-it-was-there/">have previously gone out of their way to avoid using the term</a>, instead preferring the use of “Machine Learning” to describe the technologies that underpin the experience, but if you look around iOS it is omnipresent.</p>
<img src="https://www.technicalchops.com/images/ios-ai-examples.png" alt="Screenshots of AI in use on an iPhone">
<p>Features that rely on Artificial Intelligence have been integrated throughout iOS: Cameras produce a quality of image far beyond what is achievable solely with the tiny optics fitted to the phone. Messaging has translation baked in. And swiping up in Photos gives options to look up the Artwork (or Landmark, Plant or Pet) shown in the image.</p>
<p>The approach that Apple takes is quite different from that of other manufacturers – these improvements aren’t flagship features, but relatively subtle improvements to the overall product experience that work towards that overarching goal of improving the customer experience.</p>
<h2 id="generative-hype">Generative Hype</h2>
<p>On Thursday, during an earnings call, Tim Cook addressed the question of AI directly, stating:</p>
<blockquote>
<p>As we look ahead, we will continue to invest in these and other technologies that will shape the future. That includes artificial intelligence where we continue to spend a tremendous amount of time and effort, and we’re excited to share the details of our ongoing work in that space later this year.</p>
</blockquote>
<blockquote>
<p>Let me just say that I think there’s a huge opportunity for Apple with Gen AI and AI, without getting into more details and getting out in front of myself.</p>
</blockquote>
<p>This is the most direct indication that the company is looking to bring more generative AI features to iOS. What form that will take is still speculation, but we can assume that one area that will be addressed is Siri.</p>
<h2 id="what-is-siri">What is Siri?</h2>
<p>To most, Siri is Apple’s slightly confused voice assistant.</p>
<p>Whilst this part of Siri desperately needs improvement in the context of the other services that users have become familiar with, I think that it’s unlikely that we’ll see Apple release a chatbot with the freedom that GPT (or Bing) had. I just can’t see a world where Apple allows Siri to <a href="https://stratechery.com/2023/from-bing-to-sydney-search-as-distraction-sentient-ai/">tell users they are not good people</a>.</p>
<p>Apple has an additional challenge in their overall approach: Their privacy-focused stance has seen more and more of their ML tasks performed on-device, and Apple’s long-term investment in the dedicated Neural Engine core (first introduced in 2017) has demonstrated that their strategy is very much focused on doing as much as possible without leaving the device. This results in some limitations in both the size and quality of the model that underpins Siri – what ChatGPT achieves running in a Microsoft data centre, Siri needs to within the phone.</p>
<p>The slightly lacklustre voice assistant <em>isn’t</em> Siri though. Voice is simply one of the interfaces that allows users to interact <em>with</em> Siri. Look for Siri elsewhere and you will start to see that Apple considers Siri to be almost everything that is powered by AI.</p>
<p>When widgets were introduced in iOS 14, Apple included one particular widget which I think hints at the direction of Apple’s eventual longer-term AI integration: Siri Suggestions.</p>
<p>The widget is actually two types of widget; a curated selection of 8 apps that change based on context and location, and a set of suggested actions based on what Siri anticipates you will want to do within apps, again based on your context. Whilst I think both are brilliant, and I use both types on my own home screen, it is the second that I think gives the best indication of where Apple’s AI strategy is heading.</p>
<img src="https://www.technicalchops.com/images/siri-suggestions.png" alt="Screenshots of Siri Suggestions">
<p>Apple provides the ability for apps to expose “Activities” to the wider operating system. Whether that is sending a message to a specific friend, booking a taxi or playing the next episode of a show, each activity is available to the widget without needing to go into the app to find it.</p>
<p>Within the widget, Siri then presents what it thinks are the most relevant activities for a given time and place. Arrive at a restaurant at 8pm and don’t look at your phone for two hours? Don’t be surprised if the top suggestion when you do is to book a taxi back home. Usually call your parents at 7pm on a Sunday? Expect a prompt to call them to appear. The ability for Siri to combine the contextual clues that the operating system has, its data on your historical patterns and the activities available to it allows it to have the unique ability to accurately predict what you want to do.</p>
<p>The most notable element is that the focus here is on <strong>actions</strong> using apps rather than on the app itself. This returns us to the primary driver of good user experiences; helping the user to achieve their desired outcome in the easiest possible way.</p>
<h2 id="actions-activities-and-app-clips">Actions, Activities and App Clips</h2>
<p>Given that many people have spent the past decade using a smartphone, it is not uncommon to have hundreds of apps installed, most used extremely rarely. I, for some reason, still have an app for Parc Asterix installed despite last visiting for just one day nearly 4 years ago.</p>
<p>We’re moving away from the days of “<em>There’s an app for that</em>” and into the days of “W<em>hy do I have to download an app to do that?</em>”. Apple’s solution, introduced in 2020, is App Clips.</p>
<p>App Clips are a way for developers to provide access to features of their app to a user without having them download the full app. They’re often contextual – a restaurant could provide access to their menu, ordering and payment via an App Clip via a code or NFC tag on a table. In Apple’s words: <em>“App Clips can elevate quick and focused experiences for specific tasks, the moment your customer needs them.”</em></p>
<p>Whilst I’ve rarely seen App Clips used in the wild, I sense that this is another example of Apple moving pieces into place as part of their future strategy.</p>
<h2 id="fewer-apps-more-actions">Fewer Apps, more Actions</h2>
<p>By encouraging developers to provide access to specific actions or features within an installed app or via an App Clip, Apple has created an environment for their devices in which Siri can provide users with the correct action based on context, potentially without users even needing to have the app installed.</p>
<p>As Siri’s features become more powerful, I predict that Apple will start to leverage the actions more and more, potentially even handing off the interaction entirely to Siri.</p>
<img src="https://www.technicalchops.com/images/parc-asterix.png" alt="Concept screenshots of how Parc Asterix might exist if not as an app">
<p>Take the Parc Asterix app for example – my ideal user experience is that my phone knows when I’ve arrived, checks my emails for any tickets that I already have and presents the option to buy them natively (no downloading apps) when I don’t. When I’m inside the park, I want it to provide me with easy access to an App Clip which contains a map and ride waiting times. But then I want to be able to leave and not have yet another app that won’t be used for years.</p>
<h2 id="apples-headstart">Apple’s headstart</h2>
<p>It’s easy to point at Siri’s chat functionality and suggest that Apple is falling behind, but I think the reality is quite different. Apple has spent almost a decade building AI tools that seamlessly integrate with the operating system. They have the greatest level of access to information possible because, for most people, our lives live within our phones. I want to see Apple leveraging that and integrating AI throughout the OS to work for me and make my life that much easier.</p>
<p>Where the likes of rabbit have been working on Large Action Models for the past year, Apple has been at it for a decade.</p>
<p>I do hope that Siri’s chat functionality gets a lift this year, but I don’t think that should be the focus, I want a device that anticipates me and understands what I need to make my life easier. Apple, more than anyone, is able to deliver on that.</p>
<p><em>Originally published at <a href="https://www.technicalchops.com/articles/apple-and-ai">technicalchops.com</a>.</em></p>]]></content:encoded>
      <pubDate>Thu, 01 Feb 2024 00:00:00 GMT</pubDate>
      <author>simon@minton.co (Simon Minton)</author>
      <category>AI</category>
      <category>Assistants</category>
      <category>Agents</category>
      <category>Mobile</category>
      <category>Long Term Effects</category>
      <category>LLMs</category>
      <category>Apple</category>
    </item>
    <item>
      <title>How hard is being an assistant anyway?</title>
      <link>https://www.technicalchops.com/articles/how-hard-is-being-an-assistant-anyway</link>
      <guid isPermaLink="true">https://www.technicalchops.com/articles/how-hard-is-being-an-assistant-anyway</guid>
      <description>What it takes to be a good AI Assistant</description>
      <content:encoded><![CDATA[<p>A year ago, Ben Thompson made it clear that he considers <a href="https://stratechery.com/2023/ai-and-the-big-five/">AI to be an entirely new epoch in technology</a>. One of the coolest things about new epochs is that people try out new ideas without looking silly. No one knows exactly what the new paradigm is going to look like, so everything is fair game.</p>
<p>The devices we carry every day have pretty much not changed for 15 years now. Ask a tweenager what a phone looks like and in their mind, it has likely <em>always</em> been a flat slab of plastic, metal and glass.</p>
<p>There are attempts to bring new AI powered devices that are accessories to phones to the fore – I wrote about the Ray-Ban Meta glasses <a href="https://www.technicalchops.com/articles/ray-ban-meta-glasses/">last year</a> – but it has taken until the past couple of months to see devices emerge that are clearly only possible within this new era.</p>
<p>Enter the <a href="https://hu.ma.ne/aipin">Humane Ai Pin</a> and the <a href="https://www.rabbit.tech/">rabbit r1</a>.</p>
<img src="https://www.technicalchops.com/images/rabbit-and-humane.png" alt="An illustration of the rabbit r1 and the humane Ai Pin devices">
<p>Both are standalone devices. Both have a camera and relatively few buttons. Both sport very basic screens (one a laser projector!⚡️) for displaying information to the user rather than for interaction. And both are controlled, primarily, by the user’s voice interactions with an AI assistant.</p>
<p>In theory, these are devices that you can ask to perform just about any task and they’ll just figure out how to get it done.</p>
<p>As a user, this sounds like the ideal scenario, the ultimate user experience. Issue a simple request and have it fulfilled without further interaction. The dream, like an ever-present perfect human assistant.</p>
<h2 id="the-perfect-assistant">The Perfect Assistant</h2>
<p>But let’s put aside the technology for a moment and figure out what we would expect of that perfect human assistant.</p>
<p>For the sake of this thought experiment they are invisible, always there and entirely trustworthy. Because we trust them, we would give them access to absolutely everything; email, calendar, messages, bank accounts… nothing is off limits. Why? Because they will be more effective if they have all of the same information and tools that we do.</p>
<p>So, with all of that at their disposal, the assistant should be able to solve tasks with the context of the rest of my life and their experience of previous tasks to draw on.</p>
<h3 id="simple-tasks-that-are-actually-quite-complicated">Simple Tasks That Are Actually Quite Complicated</h3>
<p>We’ll start with an easy task: “<em>Book me a taxi to my next meeting</em>”</p>
<p>The assistant knows where you currently are, so they know where to arrange the pickup from. And they have access to your calendar, so they know where the next meeting is, and what time it is. They can also look on Google Maps to check the traffic and make sure that you’ll be there on time. They know that you have an account with FreeNow, and prefer to take black cabs when you’re travelling for work. And so, when you ask them to book a taxi, they can do so relatively easily, and you will get exactly what you need.</p>
<img src="https://www.technicalchops.com/images/simple-task.png" alt="A graphic detailing the parts of the simple task of booking a taxi">
<p>Exclude one of those pieces of information though, and you will not necessarily end up with the desired result. And that’s for a relatively straightforward request.</p>
<p>When you make the request more complex, the level of information and the variety of subtasks required becomes huge. “<em>I need to meet with Customer X at their office next Wednesday</em>”. If you’re in New York and your customer is in Austen, TX, there are flights to be arranged, hotels to be booked and transfers in between, not to mention the diary management that occurs.</p>
<img src="https://www.technicalchops.com/images/complex-task.png" alt="A graphic detailing the parts of the complex task of booking a trip">
<p>These are also pretty normal requests of an assistant - things that happen the world-over every day but which are, when broken down, <em>incredibly</em> complex and made up of various interdependent subtasks. Each of them is important and if one of the subtasks fails, the entire task can go wrong.</p>
<p>The perfect assistant though, would be able to handle all of this without breaking a sweat, and we would rely on them because we trust in one of the hardest human traits to replicate: their <strong>judgement</strong>.</p>
<h2 id="enter-the-ai-ssistant">Enter the AI-ssistant</h2>
<p>The <a href="https://www.youtube.com/watch?v=22wlLy7hKP4">inferred promise in the rabbit r1 keynote</a> is that you will be able to say <a href="https://youtu.be/22wlLy7hKP4?t=864">“<em>I want to go to London around 30th January - 7th February, two adults and one child with non-stop flights and a cool SUV</em>”</a> and it will be able to plan, arrange and <strong>book</strong> the entire trip for you.</p>
<p>This, if it is true, is remarkable, precisely because of how complex and interlinked tasks actually are. If we remove the professional elements from the above example, the sub-tasks involved in booking a trip like this and the understanding required are still huge.</p>
<p>I think the r1 is a cool concept, but the hand-wavey elements of the keynote (”<em>Confirm, confirm, confirm and it’s booked</em>”) are alarming, precisely because those are the actually hard parts. <a href="https://chat.openai.com/share/faf1f487-e5b4-4ef7-b841-bdd919b2f252">Getting ChatGPT to spit out a travel itinerary is easy</a> but actually having an AI that is able to follow through and execute <strong>properly</strong> on all of the required tasks is another matter.</p>
<p>Don’t misunderstand me, I fully believe that an AI could navigate a webpage and follow the process to select and pay for things. I can see in the keynote that the r1 has the ability to connect to Expedia and would bet that it can book a hotel on the site.</p>
<p>My quandary is that when I, as an actual human™, go onto Expedia to book the above trip, I’m presented with over 300 options just for the hotels. At the top hotel, there are 7 different room types with only a couple of hundred dollars in cost difference for the entire stay between the largest and the smallest. This is already complicated before I throw in personal taste.</p>
<p>Then once you throw in flights where the option that I’d likely choose based on my personal time preferences is actually 9th on the price-ranked list (which is still within $20 of the cheapest option) and I just don’t see how the r1 is ever going to give me what I actually want. I know what that is and I know that a human assistant who has gotten to know my preferences and proclivities would likely make the same choices as I would, but that’s because we both have that very human trait of personal judgement.</p>
<p>I can see how an AI who has had access to all of my past bookings may be able to detect patterns and preferences, but I also can’t see any evidence that the r1 does have that access, or ability to learn about me personally. I won’t comment on the humane pin, but I can’t see much evidence of that there either.</p>
<p>My feeling is that a “just” good assistant, one that is just able to follow your directions and get stuff done, is actually quite hard to replicate. Combine that with the traits of a great assistant, one that can anticipate your needs, with good judgement and potentially even solve problems before you ask and we’re just at another level of complexity.</p>
<p>It’s not that I’m bearish on AI assistants as a whole but I do think that the role of being an assistant is much more complex and personal than people imagine. I can’t wait to see where we end up with daily AIs that we interact with but I can’t help but feel that an assistant in this manner just isn’t it. Yes, help me sort through the cruft of those 300 hotels, but I don’t think I’ll trust an AI to make the call for me in the same way as I would a human, at least not anytime soon.</p>
<p><em>Originally published at <a href="https://www.technicalchops.com/articles/how-hard-is-being-an-assistant-anyway">technicalchops.com</a>.</em></p>]]></content:encoded>
      <pubDate>Tue, 23 Jan 2024 00:00:00 GMT</pubDate>
      <author>simon@minton.co (Simon Minton)</author>
      <category>AI</category>
      <category>Assistants</category>
      <category>Agents</category>
      <category>Mobile</category>
      <category>Long Term Effects</category>
      <category>LLMs</category>
    </item>
    <item>
      <title>Sebastifact: A fact machine for 7 year olds</title>
      <link>https://www.technicalchops.com/articles/sebastifact-llm-facts-for-a-seven-year-old</link>
      <guid isPermaLink="true">https://www.technicalchops.com/articles/sebastifact-llm-facts-for-a-seven-year-old</guid>
      <description>Using GPT to create a fact machine</description>
      <content:encoded><![CDATA[<p>My son is 7, an age where he is becoming interested in what I do at work. He ‘gets’ the idea of apps and websites, but I wanted to put together a very simple project that we could build together so that he could see how to take an idea and turn it into a real “thing”.</p>
<p>We brainstormed some ideas – he loves writing lists of facts and finding pictures to go with them with the ambition of building an encyclopaedia, so we started work on a simple website that he could type the name of a historical person into which would then return a set of 10 facts to him.</p>
<p>His design goal was pretty simple - the website should be yellow. I decided that it was probably worth sticking to focussing on the functionality, so yellow it is.</p>
<p>As is usually the case, the backend is where the action is. I wasn’t sure how to explain to him just how complex this site would have been to make just 2 years ago - the idea of entering almost <em>any</em> historical figure into a website and having it simply provide 10 facts back would have been a hugely laborious process with many years of contributions and yet, here we are in 2023 and the solution is “just plug a Large Language Model into it”. This results in a pretty easy introductory project. We talked a little about how an API works and how computers can talk to each other and give each other instructions, and then set to work writing the prompt that we wanted to use.</p>
<p>As we tested it, he started to ask if it would work for animals too. And then mythical beasts. And then countries. Seeing him working through the ideas and realising that he could widen the scope was great. This is the eventual prompt we settled on:</p>
<p><em>You are an assistant to a family. Please provide responses in HTML. The User will provide a Historical Person, Country, Animal or Mythical Beast. Please provide 10 facts appropriate for a 7 year old. If the user provided name is not a Historical Person who is dead, a Country, an Animal or Mythical Beast please respond with an error message and do not respond with facts.</em></p>
<p>By asking the LLM to provide responses in HTML we offloaded the task of formatting the output and GPT-3.5 Turbo is pretty good at providing actual HTML. I haven’t seen any issues with it yet. By instructing it to make sure the facts are appropriate for a 7-year-old, the tone of the facts changed and we got facts that were (surprise) actually interesting for him without being too pointed in their accuracy.</p>
<p>The response takes a few seconds to come back so I implemented caching on the requests - the most popular searches appear instantly. Ideally in the future, I’ll give all of the results URLs.</p>
<p>As a final bonus, I plugged in the <a href="https://unsplash.com/developers">Unsplash API</a> to return images for him. It doesn’t always work (Unsplash apparently has relatively few pictures of Plato) but for most searches, it provides a suitable image. I might consider changing that to use the Dall-e API, but I think for now this is good enough.</p>
<p>There were two takeaways from this. Working with a 7-year-old is a test in scope creep. I wanted to keep this to an afternoon activity so that it would keep his interest but of course, it could have been a much larger site if we had incorporated all of the ideas. Giving him something that he had actually built was the important goal for me and keeping the scope to something achievable whilst also feeling like something that is his own was the most important thing. The second is something that I hammer on about all the time: an LLM is a massive toolbox that can help users achieve almost anything, but there is great value in providing a User Interface that allows a user to achieve a very specific task. There is a good reason why there are so many kitchen gadgets that can basically be replaced with a single knife - the user experience of using a dedicated tool that requires less skill <em>is</em> better.</p>
<p>The site is at <a href="https://www.sebastifact.com">https://www.sebastifact.com</a></p>
<p><em>Originally published at <a href="https://www.technicalchops.com/articles/sebastifact-llm-facts-for-a-seven-year-old">technicalchops.com</a>.</em></p>]]></content:encoded>
      <pubDate>Sat, 18 Nov 2023 00:00:00 GMT</pubDate>
      <author>simon@minton.co (Simon Minton)</author>
      <category>AI</category>
      <category>LLMs</category>
      <category>Fun</category>
    </item>
    <item>
      <title>Ray-Ban Meta Glasses: Truly Wearable AI?</title>
      <link>https://www.technicalchops.com/articles/ray-ban-meta-glasses</link>
      <guid isPermaLink="true">https://www.technicalchops.com/articles/ray-ban-meta-glasses</guid>
      <description>Giving AI Eyes and Ears</description>
      <content:encoded><![CDATA[<p>I’ve been excited to get my hands on the new <a href="https://www.meta.com/gb/smart-glasses/">Ray-Ban Meta Glasses</a>, and picked up a pair yesterday.</p>
<figure>
  <img class="grayscale" src="https://www.technicalchops.com/images/rb-m-glasses.png" alt="An illustration of a Robot wearing Ray Ban Meta Sunglasses">
  <figcaption>Me, wearing my new glasses</figcaption>
</figure>
<p>The most intriguing aspect of the glasses for me is the prospect of mixed-mode AI without taking my phone out of my pocket. Meta won’t release this until probably next year, but I do have some observations on how we could get there slightly sooner.</p>
<p>Open AI released their multi-modal version of GPT-Chat about a month ago, which means that you can now speak to Chat GPT (an oddly stilted style of conversation which is still quite compelling, <a href="https://www.technicalchops.com/articles/short-thoughts-on-chatting-with-ai">I wrote about it here</a>) and send it images which it can interpret and tell you about.</p>
<p>One of the cool features that Open AI included in the voice chat version is that on iOS the conversation is treated as a “Live Activity” – that means that you can continue the conversation whilst the phone is locked or you are using other apps.</p>
<p>What this also means is that the Ray-Ban Metas do have an AI that you can talk to, in as much as any Bluetooth headphones connected to an iPhone can be used to talk to the ChatGPT bot whilst your phone is in your pocket. I’ve looked at options to have this trigger via an automation and shortcut when the glasses connect to my phone but ultimately don’t think that is very useful - I don’t want an AI listening all the time, I want to be able to trigger it when I want it. It did lead me to add an “Ask AI” shortcut to my home screen which immediately starts a voice conversation with ChatGPT which I suppose will help me to understand how useful a voice assistant actually is over time. I also had high hopes that using “Hey Siri” would be able to trigger the shortcut, which it can, but not when the phone is locked. So close and yet so far.</p>
<p>As I said above though, this feature is also something that all headphones can be used for. The grail, and ultimate reason for getting the Ray-Bans, is in letting the AI see what you can see. Given that this feature won’t be officially released until probably next year, what options do we have?</p>
<p>The solution may come in the form of another Meta product, WhatsApp. I built a simple WhatsApp bot earlier this year which allows me to conduct an ongoing conversation with the API version of GPT-4, it’s quite rudimentary but does the job. The cool thing about the decision to deeply integrate the Meta glasses with other Meta products is that you can send messages and photos on WhatsApp directly from the glasses without opening your phone. The glasses will also read out incoming messages to you. This works pretty well with the bot that I’ve built; I can send messages using the glasses and they will read back the responses. I can say to the glasses “Hey Meta, send a photo to MyBot on Whatsapp” and it will take a snap and send it straight away. The GPT-4V(ision) API hasn’t been released yet, but once it has been, then I’ll be able to send the pictures to the bot via WhatsApp and the glasses will be able to read back the response.</p>
<p>This all feels pretty convoluted though and is ultimately an attempt to hack my way around the lack of available functionality. The Meta Glasses are quite cool but they aren’t wearable AI. Yet.</p>
<p>As with many things within the space at the moment, the technology feels tantalisingly close but not quite there. The last time anything felt quite like this to me though was the dawn of the smartphone era. Playing with the glasses has made me oddly reminiscent of playing with the accelerometer in the Nokia N95… if we’re at that point with AI then the iPhone moment is just around the corner.</p>
<p><em>Originally published at <a href="https://www.technicalchops.com/articles/ray-ban-meta-glasses">technicalchops.com</a>.</em></p>]]></content:encoded>
      <pubDate>Wed, 18 Oct 2023 00:00:00 GMT</pubDate>
      <author>simon@minton.co (Simon Minton)</author>
      <category>AI</category>
      <category>Aggregators</category>
      <category>Long Term Effects</category>
      <category>LLMs</category>
    </item>
    <item>
      <title>Short thoughts on chatting with an AI</title>
      <link>https://www.technicalchops.com/articles/short-thoughts-on-chatting-with-ai</link>
      <guid isPermaLink="true">https://www.technicalchops.com/articles/short-thoughts-on-chatting-with-ai</guid>
      <description>Multi-modal AI is here</description>
      <content:encoded><![CDATA[<p>Open AI released multi-modal AI a couple of weeks ago and it has been slowly making its way into the ChatGPT app. It is quite disconcertingly brilliant.</p>
<figure>
  <img class="grayscale" src="https://www.technicalchops.com/images/conversation.png" alt="An illustration of a Robot and a Human having a conversation.">
  <figcaption>An accurate depiction of the process</figcaption>
</figure>
<p>Conversation is a funny thing. Reading podcast transcripts can be quite nightmarish - we don’t realise how much the spoken word, especially during conversations, meanders and is peppered with hesitation, deviation and repetition until we see it written down. When we’re speaking though, these unnecessary additions make conversation human and enjoyable. When we’re listening, we often don’t realise how much we are actually playing an active role in doing so–in person, it’s the facial expressions and nods which encourage the speaker to continue; on the phone, the short acknowledgements that let a partner know you’re still there and listening. I often speak to a friend on the phone who mutes when they’re not speaking and the experience is fine, but the silence is slightly off-putting.</p>
<p>And so to the experience of chatting with an AI. It’s brilliant, in as much as it actually feels as though you are having something of a conversation. The responses aren’t the same as the ones you would receive by directly typing the same words into Chat GPT - they’ve clearly thought about the fact that spoken conversation is different. There is surprisingly little lag in the response. You don’t say your piece and then wait for 10 seconds for it to process; the AI responds in a couple of seconds, almost straight away once it’s heard a long enough pause. The quality of the AI is fantastic - it’s using GPT-4 which is about as state-of-the-art as it can get, and the voices, whilst not human, are surprisingly great.</p>
<p>However.</p>
<p>The entire experience is disconcerting because of how precise it is. There is no room for you to take long pauses while you think mid-sentence, or rephrase as you talk. There is absolute silence when you are talking which causes you to look down at the screen to make sure it’s still working. The responses are often long and apparently deeply thought through, but they often end with a question, rather than just being an open-ended response to work from. I’m looking forward to having an AI conversational partner, but I want it to help me tease out ideas, not necessarily give me fully formed AI thoughts on a subject. I want it to say “yes” whilst I’m speaking for no apparent reason other than to encourage me to keep talking through the idea. I want it to meander and bring in new unrelated but tangential ideas. Ultimately, I guess I want it to be a little more human.</p>
<p><em>Originally published at <a href="https://www.technicalchops.com/articles/short-thoughts-on-chatting-with-ai">technicalchops.com</a>.</em></p>]]></content:encoded>
      <pubDate>Tue, 10 Oct 2023 00:00:00 GMT</pubDate>
      <author>simon@minton.co (Simon Minton)</author>
      <category>AI</category>
      <category>Voice</category>
      <category>Multi-Modal</category>
      <category>LLMs</category>
    </item>
    <item>
      <title>OK Computer: 21st Century Sounds</title>
      <link>https://www.technicalchops.com/articles/ok-computer-21st-century-sounds</link>
      <guid isPermaLink="true">https://www.technicalchops.com/articles/ok-computer-21st-century-sounds</guid>
      <description>Musical curation and creation, Spotify and AI</description>
      <content:encoded><![CDATA[<h2 id="musical-discovery">Musical Discovery</h2>
<p>For much of the latter half of the 20th century, new music discovery went something like this. An artist would make a song and they’d send demo tapes out to record companies and radio stations. They’d play to dimly lit bars and clubs, hoping that an A&#x26;R impresario lurked in the crowd. If they were lucky, a DJ might listen to their demo and would play it live. Perhaps someone would record it and start bootlegging tapes. These contraband tapes would be passed around and listened to by teenagers gathered in bedrooms. If all went well, the artist’s popularity would grow. They’d be signed, played more on the radio and do bigger shows. Fans and soon-to-be fans would go to record stores to listen to the new releases and buy the music on vinyl, tape or CD. The record shops would make money, the musicians would make money, the record companies would make more money.</p>
<figure>
  <img class="grayscale" src="https://www.technicalchops.com/images/dj.png" alt="An illustration of a Robot DJing.">
  <figcaption>R2-DJ</figcaption>
</figure>
<p>This began to shift in the early naughties, driven, as so much was, by the emergence of the internet. The newfound ability to rip CDs and transform tracks into easily shareable mp3s on the likes of Napster rendered the entire world’s music repertoire available gratis to eager ears. For those that preferred their music to come without the lawbreaking, the iTunes store and others made purchasing it just about as easy. MP3 players and the iPod made it <a href="https://www.youtube.com/watch?v=6SUJNspeux8">effortless to carry 1000 songs in your pocket</a>. The days of physically owning your music were all but over in the space of only a few short years.</p>
<p>Despite the music industry’s hope that <a href="https://www.wired.com/2002/05/the-day-the-napster-died/">killing Napster</a> would stem the rising tide, the death of the platform only resulted in more alternatives appearing. It turned out that people liked having instant access to all music for pretty much free. Music discovery underwent a transformation. To acquire a song, one simply had to search for it, and within minutes, it was yours—provided Kazaa or uTorrent were operational and your parents didn’t pick up the phone and break the connection. Online forums teemed with enthusiasts discussing new musical revelations and leaks, offering nearly everything and anything you desired, all for free.</p>
<p>Music was no longer scarce; there were effectively infinite copies of every single song in the world to which anyone could have immediate access. Gone were the days of friends passing around tapes or lingering in record stores. The social aspect of music discovery shifted from smoky bars, intimate bedrooms, and record emporiums to the virtual amphitheaters of online forums, Facebook threads, and text message exchanges.</p>
<p>The big problem with all of this of course was that it was all quite illegal.</p>
<p>In 2006, Daniel Ek was working at uTorrent. He could see that users wouldn’t stop pirating music and that the attempts by the Music Industry to thwart sharing were doomed to failure. He “<em>realized that you can never legislate away from piracy. Laws can definitely help, but it doesn’t take away the problem. The only way to solve the problem was to create a service that was better than piracy and at the same time compensates the music industry – that gave us Spotify.”</em></p>
<h2 id="musical-curation">Musical Curation</h2>
<p>Spotify launched with the simple headline: A world of music. Instant, simple and free.</p>
<p>By 2023 it has over 500 million users.</p>
<p>For many music fans, playlists took center stage, with enthusiasts investing hours <a href="https://techcrunch.com/tag/sharemyplaylists/">trawling the internet for them</a>. Spotify introduced a feature to see what your friends were listening to via Facebook and then to directly share playlists with others. Then they made playlists searchable. That killed off sites like sharemyplaylist, but meant that when I needed three hours of Persian Wedding songs, all I had to do was hit the search bar and appear to be intimately familiar with the works of <a href="https://spotify.link/sNrxiMvSCDb">Hayedeh</a> and <a href="https://spotify.link/BLVS9KDSCDb">Dariush</a> for my then soon-to-be in-laws.</p>
<p>In 2015 <a href="https://artists.spotify.com/en/discovery-mode">Spotify launched Discover</a>, a dynamic playlist which introduced users to tracks that were similar to what the listener had played recently. It was remarkably good. The social aspect of music discovery was being lost but it was replaced with an automaton that did the job exceptionally well, even if the results were sometimes corrupted by the false signal of repeated plays of Baby Shark.</p>
<p>What was more subtle about what had been happening throughout this period was that the way people consumed music was changing. We had progressed from music discovery as a purposeful act to one in which it was an every day occurrence. Background music had always existed, but it was via the radio, or compilations. This was personal. The value of the music itself transformed. The ability to have a consistent soundtrack playing, at home, at work, in the car or as you made your way through every day life, meant that listeners weren’t necessarily concerned about the specific artists that were playing, they had become more interested in the general ambience of that ever-present background music. Listeners still certainly relished the release of the new Taylor Swift album, but they also listened to music that they didn’t know more easily and without ever inquiring as to who the artist was, simply because it fit within the soundtrack of their lives.</p>
<p>The Discover feature was one of Spotify’s first public forays into personalised music curation using machine learning. The success of the project led to more experiments. It turned out that people <em>loved</em> the feature.</p>
<p>Spotify in 2023 is remarkable. When I want to run to rock music, the tempo and enthusiasm of the suggested playlist is exactly right. When I want to focus on work, the beats are mellow and unobtrusive. The playlists “picked for me” change daily, powered by AI. I still create my own playlists, but the experience is now akin to using ChatGPT. I add a few songs to set a general mood and Spotify offers up suggestions that match the general vibe. Prior to a recent trip, I created a playlist called “Italy Background Music”, which Spotify duly filled with tracks I wouldn’t have had the first idea about where to find. They were exactly what I was looking for.</p>
<p>Curation and general discovery, it seems, have been broadly solved by Spotify.</p>
<h2 id="musical-creation">Musical Creation</h2>
<p>I’ve become accustomed to hearing tracks that I’ve never before heard and wouldn’t have the first idea about the artists of. Occasionally, I’ve tapped through on an unknown song and discovered that it has only had a couple of hundred thousand plays, ever. Spotify is clearly drawing on the entire breadth of artists within its library to match my musical preferences. Or is it?</p>
<p>In 2016, <a href="https://www.musicbusinessworldwide.com/spotify-is-creating-its-own-recordings-and-putting-them-on-playlists">Music Business Worldwide published an article</a> stating:</p>
<blockquote>
<p>Spotify is starting to make its own records.</p>
<p>Multiple cast-iron sources have informed us that, in recent months, Daniel Ek‘s company has been paying producers to create tracks within specific musical guidelines.</p>
</blockquote>
<p>By introducing its own music into the (literal) mix, for which it has paid a producer a flat fee, has added to the platform under a false name and then surfaced to listeners via its AI curated playlists, the platform is solving two issues that it considers important: from a user’s perspective, more music that fits their desired soundscape is a good thing. From Spotify’s perspective, having the ability to add, say, a 3 minute track of in-house music (which they don’t need to pay royalties on) to every hour of listening means that their cost for that hour is reduced by 5%. The losers in this case are the artists, who would otherwise have earned from that three minute play.</p>
<p>There is nothing that says that Spotify can’t do this though, it’s their platform and <a href="https://time.com/3554468/why-taylor-swift-spotify/">even when big artists have removed their music from the platform in protest</a> at the company’s policies or actions, they’ve quietly reappeared within months, such is the value of the service<sup><a href="#user-content-fn-1" id="user-content-fnref-1" data-footnote-ref="" aria-describedby="footnote-label">1</a></sup>.</p>
<p>In the above 2016 article, it is clear that the firm was paying actual producers to make the music. In 2023, the landscape is likely quite different. AI has advanced to such a point where the beats, melodies and riffs of jazz, trip hop and other non-vocal music can be quite easily produced by a well trained AI. There are dozens of <a href="https://lofigenerator.com/">sites that algorithmically generate lo-fi background music</a>. If Spotify isn’t already adding tracks generated by AI that perfectly match a given vibe, especially within those non-vocal musical genres, then it is at least experimenting with it. The prize is too large to not. In 2021, the company <a href="https://www.notion.so/Spotify-da64ef456aa0446698bb707ac0bd0bd7?pvs=21">paid more than $7bn to rights holders</a>. At 5%, that’s a nice $350m to find down the back of the AI sofa.</p>
<h2 id="licensing">Licensing</h2>
<p>Where this leads in my mind is something sort of entirely new.</p>
<p>Whilst vocal-less (can we really call it instrumental?) music is the easiest use of the technology, we saw earlier this year as there was a brief explosion of AI generated tracks from creators using AI voice models that imitated the likes of <a href="https://www.youtube.com/watch?v=HLsB-hVV024">Drake</a> and <a href="https://twitter.com/rpnickson/status/1639813074176679938">Kanye</a>. Whilst these tracks weren’t perfect, they showed an early preview of technology that will change the face of music. <a href="https://huggingface.co/docs/transformers/community">The Hugging Face community</a> is full of models of popular artists which can replicate the sound of a given singer or rapper and it is evident improvements move at a rapid clip, with some now indistinguishable from the original artist.</p>
<p>Licensing of brands exists broadly in most other industries. In fashion <a href="https://www.joincolossus.com/episodes/69912643/senra-ralph-lauren-the-man-behind-the-mystique">it saved Ralph Lauren</a> (although <a href="https://www.wsj.com/articles/burberry-revamps-its-image-in-japan-1439547804">nearly killed Burberry</a>). It famously <a href="https://www.acquired.fm/episodes/nike">turned Nike from pure sportswear to casual fashion mega-brand</a>. Could we see the emergence of the artist as a brand? The potential for artists to either directly license their musical likeness to a given platform or to allow producers to use an authorised AI model of their voice to create tracks which they, or their team, would have final sign off on could allow vocalists to extend their reach drastically.</p>
<p>Whilst the last idea might sound fanciful, there are artists who already draw on the online community. One of the DJ/Producer Deadmau5’s biggest tracks was the <a href="https://www.youtube.com/watch?v=iqurYVWg7-g">result of a fan sending him a demo of a vocal mix via twitter</a> for a track that Deadmau5 had produced the previous day in a livestream.</p>
<p>We’ve also seen a rash of artists selling their music rights–will the future see those artists who reach the end of their careers sell their “official” AI model to allow them and their families to earn in perpetuity? It’s been proven repeatedly that those artists who adapt to the changing world are the ones that succeed, but this is something entirely new.</p>
<p>What seems certain however is that the music that we listen to in the coming years will be picked for us by machines and at least partially created using AI.</p>
<section data-footnotes="" class="footnotes"><h2 class="sr-only" id="footnote-label">Footnotes</h2>
<ol>
<li id="user-content-fn-1">
<p>There is a good argument that Taylor’s reasoning for removing her music wasn’t entirely to do with this. <a href="#user-content-fnref-1" data-footnote-backref="" class="data-footnote-backref" aria-label="Back to content">↩</a></p>
</li>
</ol>
</section>
<p><em>Originally published at <a href="https://www.technicalchops.com/articles/ok-computer-21st-century-sounds">technicalchops.com</a>.</em></p>]]></content:encoded>
      <pubDate>Fri, 15 Sep 2023 00:00:00 GMT</pubDate>
      <author>simon@minton.co (Simon Minton)</author>
      <category>Aggregators</category>
      <category>Music</category>
      <category>Personalisation</category>
      <category>Long Term Effects</category>
      <category>Media</category>
      <category>Licensing</category>
    </item>
    <item>
      <title>It’s about to get very noisy</title>
      <link>https://www.technicalchops.com/articles/its-about-to-get-very-noisy</link>
      <guid isPermaLink="true">https://www.technicalchops.com/articles/its-about-to-get-very-noisy</guid>
      <description>Large Language Models (LLMs), content creation and filtering signal from noise</description>
      <content:encoded><![CDATA[<p>The internet is about to get a lot noisier, thanks to the rise of Large Language Models (LLMs) like GPT.</p>
<figure>
  <img class="grayscale" src="https://www.technicalchops.com/images/typing.png" alt="An illustration of a Robot Monkeys typing on typewriters">
</figure>
<h3 id="the-creators-of-the-internet">The Creators of The Internet</h3>
<p>40 million people sounds like <em>a lot</em> of people. It’s roughly the population of Canada. And in 1995, it was the population of the entire internet. But today, today the population of the internet is about 5.2 billion. That is quite a lot more people. Most of those people are consumers. They use the internet without adding anything to it; this is fine, it is good in fact. They watch Netflix, chat on WhatsApp and scroll on TikTok. They might tweet on Twitter but, since the great Elon-ification of that place, it appears that even fewer of them are even doing that. These people used to be called ‘<a href="https://en.wikipedia.org/wiki/Lurker">lurkers</a>’ on forums. What they aren’t doing is creating websites or trying to get you to buy something or writing blogs like this one. For our purposes we can say that these people are being very quiet indeed.</p>
<p>But then there are a <a href="https://en.wikipedia.org/wiki/1%25_rule">much smaller group of people</a> who, like me, see the internet as the place that they can make a bit of noise and generally conduct some form of creativity, be it in running a business or creating reels for consumers to scroll or cobbling their thoughts into a 1000 word article. This is also a good thing - consumers can’t consume if no one is creating stuff for them to consume.</p>
<h3 id="signal-vs-noise">Signal vs Noise</h3>
<p>Most of the stuff that is being created is pretty average; it’s just background noise, a light hum. A few people might consume it, but it’s receiving 27 likes on Instagram and isn’t really getting in anyone’s way. It’s the background music that plays in shops and hotel lobbies that fills the uncomfortable silence. Always there, but quite pleasant and you’d probably miss it if it wasn’t there.</p>
<p>Then we have <em>the good stuff</em> - these are the pieces that go viral: the think piece that nails a topic so absolutely, the tiktok that gets a bajillion views, the hot take that turns out to be absolute fire, the stack overflow answer that comes like an angel of the night to salve your programming pain. This is the sweet melody of signal. We like signal, and we spend most of our consuming time wading through the mire of noise looking for it. Sure the background noise is a bit irritating, but it’s relatively easy to find <em>the good stuff</em> once you know where and how to look for it. Google is pretty good at finding what we want. We’ve built aggregators like reddit or hacker news, where people can come and say “Hey everyone! Look, I found some gold!” and then other people can upvote it if it’s actually gold, or downvote it to oblivion if it’s not actually good.</p>
<p>All of this seems sort of fine. Every year more noise is created, but so too is signal. The good thing about this is that an individual creator could only create so much noise. Automated content was pretty obviously automated content, and even if Google didn’t manage to filter it out for you first, once you started reading it, it was pretty clear that no human had set eyes on it during the creation process. We become quickly attuned to that particular lack of harmony and click the back button, which also signalled to Google that the content wasn’t actually helpful, meaning the next person searching for that particular query would be less likely to end up seeing that page.</p>
<p>The problem we’re facing now though, is that creators have been given a whole new kind of instrument to play. It’s not that this instrument is louder or more dominant that the others, it’s that it can create an almost infinite amount of songs with barely any human input at all. And they all sound pretty great. LLMs are <em>really</em> good at creating noise. It’s not just that they can create an ungodly amount of content (basically for free) but differentiating that content from human generated content is, by design, almost impossible. Where a content creator could once have put out a few decent quality articles a day, they can now put out thousands. The ratio of noise to signal is about to dramatically shift.</p>
<p><em>Originally published at <a href="https://www.technicalchops.com/articles/its-about-to-get-very-noisy">technicalchops.com</a>.</em></p>]]></content:encoded>
      <pubDate>Thu, 15 Dec 2022 00:00:00 GMT</pubDate>
      <author>simon@minton.co (Simon Minton)</author>
      <category>AI</category>
      <category>Marketing</category>
      <category>Long Term Effects</category>
      <category>LLMs</category>
    </item>
    <item>
      <title>Webinars and the Evolution of Playbox</title>
      <link>https://www.technicalchops.com/articles/webinars-and-the-evolution-of-playbox</link>
      <guid isPermaLink="true">https://www.technicalchops.com/articles/webinars-and-the-evolution-of-playbox</guid>
      <description>Playbox: My first foray into an AI product</description>
      <content:encoded><![CDATA[<p>In early 2020, as the world adjusted to a new normal, I found myself inundated with invitations to webinars. Despite their promise of convenience and flexibility, I couldn’t shake the image of low-ceilinged conference rooms at a Holiday Inn, time-share salesmen, and the lingering smell of stale coffee. My aversion to webinars, though, seemed irrational. After all, webinars could be attended from the comfort of home, with self-brewed coffee, and the ever-tempting option to click the X and leave whenever they morphed into disguised sales pitches. No accusatory stares included.</p>
<p>Yet, as the pandemic stretched on, attending webinars became a daily ritual. The more I participated, the clearer it became that my frustration stemmed not just from the content but from the format itself. Companies would often plan an hour-long session, with 45 minutes dedicated to presentations and a mere 15 minutes for questions at the end. The structure seemed arbitrary, with many hosts struggling to fill the time meaningfully. The low information density and the “questions at the end” format meant I often lost track of my queries or interest by the time my turn came.</p>
<p>Determined to find a solution, I envisioned a more efficient webinar experience. Assuming the content was relevant, my ideal setup included:</p>
<ul>
<li>A searchable transcript, complete with images of key slides</li>
<li>Bullet points highlighting the top takeaways</li>
<li>The ability to watch at 1.5x speed or higher</li>
<li>An option to listen to the webinar as a podcast</li>
</ul>
<p>Achieving this, however, posed a significant challenge for production teams, requiring considerable effort to create transcripts, summarize content, and upload podcasts. Enter GPT-3. With its release, I saw an opportunity to transform how webinars were consumed.</p>
<p>This marked the birth of Playbox (née Foyer), driven by my vision that any input should become any output, allowing users to consume content in their preferred format.</p>
<p>We utilized <a href="https://www.rev.ai/">rev.ai</a> to achieve 98% accurate transcripts, developed a sophisticated transcript editor linking words to video timestamps, and allowed users to insert video frames to display slides. GPT-3 helped generate summaries and key takeaways, enabling easy creation of social media snippets and videos and podcast-ready audio tracks with chapters.</p>
<p>However, understanding the motivations behind companies hosting webinars proved complex. Typically part of a broader marketing strategy, webinars often aimed to gather potential customer details. While we addressed this by allowing content to be gated behind email collection forms, the core issue was that many companies used webinars not just to inform but to engage potential customers post-event. By enabling users to skip the live experience, Playbox was seen as disrupting this lead generation flow, despite evidence that our approach led to a gradual increase in leads over time.</p>
<p>In hindsight, the timing for Playbox was off. During a period when webinars were a primary touchpoint, offering a tool that encouraged asynchronous interaction conflicted with companies’ immediate needs. Yet, I remain convinced that a consumer-centric approach to content consumption is the future. Companies that embrace this shift, making their content accessible in all formats, will ultimately gain more control and foster deeper connections with their audience.</p>
<p>While Playbox’s initial reception highlighted the challenges of disrupting entrenched practices, the principles behind it hold promise. As technology and consumer preferences evolve, the idea of flexible, on-demand content consumption, in the format that they want to consume, will become increasingly relevant. Companies that adapt to this new landscape will not only survive but thrive, meeting their audience on their terms and fostering lasting engagement.</p>
<p><em>Originally published at <a href="https://www.technicalchops.com/articles/webinars-and-the-evolution-of-playbox">technicalchops.com</a>.</em></p>]]></content:encoded>
      <pubDate>Thu, 15 Dec 2022 00:00:00 GMT</pubDate>
      <author>simon@minton.co (Simon Minton)</author>
      <category>AI</category>
      <category>Webinars</category>
      <category>Media</category>
      <category>LLMs</category>
    </item>
    <item>
      <title>Welcome to Technical Chops</title>
      <link>https://www.technicalchops.com/articles/welcome-to-technical-chops</link>
      <guid isPermaLink="true">https://www.technicalchops.com/articles/welcome-to-technical-chops</guid>
      <description>Writing on the internet</description>
      <content:encoded><![CDATA[<p>One of the most exciting elements of working within <em>the tech industry</em> is that there is always lot of ✨new✨.</p>
<p>New ideas, new products, new words, new technologies. New things are good. New things are shiny. But which new things are going to change the world? Which technologies are going to fundamentally change how we interact with the world? How do we differentiate between hype and practical utility? And what does this all mean for the vast majority of the world?</p>
<p>As a non technical person, and even as a technical person, it is increasingly difficult to differentiate between the simply new and shiny and the new, shiny and potentially world changing. I’m looking forward to exploring how new technologies could change the world, and what it means for our next rotation.</p>
<p>Over the past few years I’ve been paying particular attention to:</p>
<ul>
<li>Open Source Software and its wider adoption and integration</li>
<li>Artificial Intelligence</li>
<li>The evolution of software development and the speed with which the hard becomes easy</li>
<li>The deep integration of technology into every facet of our lives</li>
</ul>
<p>I’m committing to one article a week exploring these topics and more. Follow along here or at <a href="https://www.twitter.com/technicalchops">@technicalchops</a> on Twitter.</p>
<p><em>Originally published at <a href="https://www.technicalchops.com/articles/welcome-to-technical-chops">technicalchops.com</a>.</em></p>]]></content:encoded>
      <pubDate>Thu, 08 Dec 2022 00:00:00 GMT</pubDate>
      <author>simon@minton.co (Simon Minton)</author>

    </item>
  </channel>
</rss>