I floated a theory of mine to Atlantic writer Charlie Warzel on this week’s episode of Panic World that content moderation, as we’ve understood, it effectively ended on January 6th, 2021. You can listen to the whole episode here, but the way I look at it is that the Insurrection was the first time Americans could truly see the radicalizing effects of algorithmic platforms like Facebook and YouTube that other parts of the world, particularly the Global South, had dealt with for years. A moment of political violence Silicon Valley could no longer ignore or obfuscate the way it had with similar incidents in countries like Myanmar, India, Ethiopia, or Brazil. And once faced with the cold, hard truth of what their platforms had been facilitating, companies like Google and Meta, at least internally, accepted that they would never be able to moderate them at scale. And so they just stopped.

After 2021, the major tech platforms we’ve relied on since the 2010s could no longer pretend that they would ever be able to properly manage the amount of users, the amount of content, the amount of influence they “need” to exist at the size they “need” to exist at to make the amount of money they “need” to exist.

Under Zuckerberg’s new “censorship”-free plan, Meta’s social networks will immediately fill up with hatred and harassment. Which will make a fertile ground for terrorism and extremism. Scams and spam will clog comments and direct messages. And illicit content, like non-consensual sexual material, will proliferate in private corners of networks like group messages and private Groups. Algorithms will mindlessly spread this slop, boosted by the loudest, dumbest, most reactionary users on the platform, helping it evolve and metastasize into darker, stickier social movements. And the network will effectively break down. But Meta is betting that the average user won’t care or notice. AI profiles will like their posts, comment on them, and even make content for them. A feedback loop of nonsense and violence. Our worst, unmoderated impulses, shared by algorithm and reaffirmed by AI. Where nothing has to be true and everything is popular.

the most advanced generalist AI models often outperform specialized models, even in the specific domains those specialized models were designed for. That means that if you want a model that can do a lot - reason over massive amounts of text, help you generate ideas, write in a non-robotic way — you want to use one of the three frontier models: GPT-4o, Gemini 1.5, or Claude 3 Opus.

Working with advanced models is more like working with a human being, a smart one that makes mistakes and has weird moods sometimes. Frontier models are more likely to do extraordinary things but are also more frustrating and often unnerving to use. Contrast this with Apple’s narrow focus on making AI get stuff done for you.

Every major AI company argues the technology will evolve further and has teased mysterious future additions to their systems. In contrast, what we are seeing from Apple is a clear and practical vision of how AI can help most users, without a lot of effort, today. In doing so, they are hiding much of the power, and quirks, of LLMs from their users. Having companies take many approaches to AI is likely to lead to faster adoption in the long term. And, as companies experiment, we will learn more about which sets of models are correct.

Quality software from independent makers is like quality food from the farmer’s market. A jar of handmade organic jam is not the same as mass-produced corn syrup-laden jam from the supermarket. Industrial fruit jam is filled with cheap ingredients and shelf stabilizers. Industrial software is filled with privacy-invasive trackers and proprietary formats. Google, Apple, and Microsoft make industrial software. Like industrial jam, industrial software has its benefits — it’s cheap, fairly reliable, widely available, and often gets the job done.

Big tech companies have the ability to make their software cheap by subsidizing costs in a variety of ways:

Google sells highly profitable advertising and makes its apps free, but you are subjected to ads and privacy-invasive tracking. Apple sells highly profitable devices and makes its apps free, but locks you into a proprietary ecosystem. Microsoft sells highly profitable enterprise contracts using a bundling strategy, and makes its apps cheap, also locking you into a proprietary ecosystem.

I’m not saying these companies are evil. But their subsidies create the illusion that all software should be cheap or free.

Independent makers of quality software go out of their way to make apps that are better for you. They take a principled approach to making tools that don’t compromise your privacy, and don’t lock you in. Independent software makers are people you can talk to. Like quality jam from the farmer’s market, you might become friends with the person who made it — they’ll listen to your suggestions and your complaints.

Big tech companies earn hundreds of billions of dollars and employ hundreds of thousands of people. When they make a new app, they can market it to their billions of customers easily. They have unbeatable leverage over the cost of developing and maintaining their apps.

But more than anything, they were able to merge with major competitors and buy out small ones. Google made one good product, search, a quarter of a century ago. That opened conduit to the capital markets that gave Google an effectively limitless budget to buy competitors.So it didn’t matter that everything Google made in-house failed — videos, social media, wifi balloons, smart cities, they couldn’t even keep an RSS Reader alive!Because they were able to buy other peoples’ companies — mobile, ad tech, videos, maps, documents, satellites, server management. Google isn’t Willy Wonka’s magic idea factory, they’re Rich Uncle Pennybags, spending other peoples’ money to buy the products they themselves are too ossified and lumbering to create.

They were able to sell goods below cost, which let the deepest-pocketed companies bankrupt their competitors, and prevent new companies from entering the market. Think of Amazon, which tried to buy diapers.com, got rejected, and then lit $100m on fire selling diapers below cost, until diapers.com went bankrupt.

When Apple reversed Office and built iWork, Microsoft just had to suck it up. In the ensuing decades, Apple — and Microsoft, Facebook, Google and other tech giants — have secured changes to law, regulation and their interpretations that make doing unto them as they did unto others radioactivelyillegal.

Tech companies can twiddle the knobs whenever they want, without explanation or transparency, and we can’t get a law passed to make them stop compulsively touching their knobs, because in the world of five giants websites each filled with screenshots of the other four, they can easily agree that these rules are bad, and they can mobilize their monopoly casino winnings to make sure they never pass.

Step one: consolidated industries eliminate competition through predatory pricing and acquisitions. Step two: tech companies play a high-speed shell-game on the back end, and use their consolidation to bigfoot any attempt to constrain their twiddling (like privacy, labor, or fair trading laws). Now we come to step thre: where tech companies embrace tech laws, laws that make it illegal to twiddle back at them, the IP laws that create felony contempt of business-model, criminalizing the adversarial interoperability, that once acted as garbage collection for enshittified, bloated, top-heavy companies, letting nimble, innovative players drain off their users, eat their lunch and dance on their graves.Put these three factors together — consolidation, unrestricted twiddling for them, a total ban on twiddling for us — and enshittification becomes inevitable.

We don’t want to wait that long for a new good internet, and we don’t have to. Because tech is different: it is universal. It is interoperable, and that means we have options we’ve never had before.Interoperability options: options that devolve control over technology from giant companies to small companies, co-ops, nonprofits, and communities of users themselves.Interop is how we seize the means of computation.

First things first: we need to limit twiddling.Pass comprehensive federal privacy laws with private right of action, meaning that you can sue if your privacy is violated, even if the local public prosecutor doesn’t think you deserve justice.End worker misclassification through the so-called gig economy, meaning that every worker is entitled to minimum wages, a safe workplace, and fair scheduling.Apply normal consumer protection standards to ecommerce platforms and search engines, banning deceptive advertising, fake reviews, and misleading search results that put fake businesses and products ahead of the best matches.

Then we need to open the walled gardens. Laws like the EU’s Digital Markets Act will force tech platforms to stand up APIs that allow new platforms to connect to them. This interop will make switching costs low. So you can leave Facebook or Twitter and go to Mastodon, Diaspora — or Bluesky or some new platform — and still exchange messages with the people you left behind, and participate in the communities that matter to you, and connect with the customers you rely on.

To make mandatory APIs work, we need to make robust interoperability preferable to behind-the-scenes fuckery, we need to align tech giants’ incentives so they encourage competition, rather than sabotaging it.

in addition to the mandatory interop that’s already coming down the pike, we need to restore the right to mod, tinker, reverse and hack these services.

If we have the right to mod existing service to restore busted API functionality, then any company that’s tempted to nerf its API has to consider the possibility that you are going to come along and scrape its site or reverse its apps to make the API work again.That means that the choice for tech giants isn’t “Keep the API and lose my discontented users or nerf the API and screw my competitors.” It’s: “Keep the API and lose my discontented users or, nerf the API and get embroiled in unquantifiable guerilla warfare against engineers who have the attackers’ advantage, meaning I have to be perfect, and they only have to find and exploit a single error I make.”

Governments should require that every tech company that sells them a product or service has to promise not to interfere with interop.That’s just prudent public administration. Lincoln insisted that every rifle-supplier for the Union army used interoperable tooling and ammo. Of course he did! “Sorry boys, war’s cancelled, our sole supplier decided not to make any more bullets.”

Every digital system procured by every level of government should come with a binding covenant not to impede interop — from the cars in government motor-pools to Google Classroom in public schools to iPhones in public agencies.

Your shareholders’ priorities are your problem. Public agencies are charged with doing the people’s business.

It’s frankly surreal that the way we keep Facebook’s partners from abusing your info is by asking Facebook to decide what is and isn’t acceptable.Remember: Cambridge Analytica was a Facebook partner. So whether you’re using an API or you’re fielding an interoperable app that relies on scraping and reversing, you will be bound by those same laws, passed by democratically accountable lawmakers in public proceedings, not by shareholder accountable corporate executives in closed-door meetings.

They’re just able to buy their way to dominance, merging with competitors, until they have the money and the unity of purpose to capture our laws, to give them the freedom to abuse us without limit, and to criminalize anything we do to defend ourselves.To stop them we need to block new merger, and unwind existing ones, limit their ability to twiddle the back end to keep their users and business customers in a constant state of confusion, and restore our ability to twiddle back, to give ourselves an internet operated by and for the people who use it: the new, good internet that is the worthy successor to the old, good internet.

Remember when tech workers dreamed of working for a big company for a few years, before striking out on their own to start their own company that would knock that tech giant over?Then that dream shrank to: work for a giant for a few years, quit, do a fake startup, get acqui-hired by your old employer, as a complicated way of getting a bonus and a promotion.Then the dream shrank further: work for a tech giant for your whole life, get free kombucha and massages on Wednesdays.And now, the dream is over. All that’s left is: work for a tech giant until they fire your ass, like those 12,000 Googlers who got fired six months after a stock buyback that would have paid their salaries for the next 27 years.

“Some day, there will be a crisis, and when crisis comes, ideas that are lying around can move from the fringe to the center in an instant.”

We still believe that there will be a separation between the “application layer” companies and foundation model providers, with model companies specializing in scale and research and application layer companies specializing in product and UI. In reality, that separation hasn’t cleanly happened yet. In fact, the most successful user-facing applications out of the gate have been vertically integrated.

We predicted that the best generative AI companies could generate a sustainable competitive advantage through a data flywheel: more usage → more data → better model → more usage. While this is still somewhat true, especially in domains with very specialized and hard-to-get data, the “data moats” are on shaky ground: the data that application companies generate does not create an insurmountable moat, and the next generations of foundation models may very well obliterate any data moats that startups generate. Rather, workflows and user networks seem to be creating more durable sources of competitive advantage.

Some of the best consumer companies have 60-65% DAU/MAU; WhatsApp’s is 85%. By contrast, generative AI apps have a median of 14% (with the notable exception of Character and the “AI companionship” category). This means that users are not finding enough value in Generative AI products to use them every day yet.

generative AI’s biggest problem is not finding use cases or demand or distribution, it is proving value. As our colleague David Cahn writes, “the $200B question is: What are you going to use all this infrastructure to do? How is it going to change people’s lives?”

I think hardcore Twitter users have rose-colored glasses about the site’s coolness. The reason for its success, if you can argue that it was ever really successful, wasn’t that it was cooler than Facebook. It was because of its proximity to power. The reason it was so popular with activists, extremists, journalists, and shitposters was because what you posted there could actually affect culture.

The thing that ties together pretty much everything that’s happened on Twitter since it launched in 2006 was the possibility that those who were not in power (or wanted more) could influence those who were.

I subscribe to the belief that internet trends are defined by a ratio of laziness to social reward. Users will always do the laziest possible thing to achieve the maximum amount clout. So, if every platform becomes either a Twitter alternative or a short-form video feed, but all with their own unique requirements for virality, users won’t make individual posts for each. They will instead shotgun blast all of them with the same posts and bet on the odds that something will breakthrough eventually. Which means everything eventually just becomes a reuploaded video or a screenshot from somewhere else.

While trying to track down the actual hyperlink to a post I found a screenshot of on a closed social network I was struck by how on an internet full of closed platforms, broken embeds, and crumbling indexes, the last reliable way to share anything is a screenshot.

the camera roll is, at this point, the real content management system of the social web. This is something that TikTok realized faster than other platforms, with their downloadable watermarked videos that have now become ubiquitous on every platform that allows video.

My theory as to why New Yorkers were so allergic to independent content creators is because for all the tedious guffawing about being a city of hustlers, most of the people who live there crave, on some level, institutional legitimacy and influencers, by definition, don’t get it or really need it. It could also just be that New Yorkers hate tourists and content creators are, in some form, permanent tourists of their own lives.

I actually think the post-COVID New York TikTok boom is already cresting. I think once these trends become calcified enough to report on, they’re already on their way out. I also don’t think Gen Z TikTokers are driving rents up, but rather documenting its rise due to other factors, like landlords being able to blame TikTok hype to jack up their rents.

Google is trying to kill the 10 blue links. Twitter is being abandoned to bots and blue ticks. There’s the junkification of Amazon and the enshittification of TikTok. Layoffs are gutting online media. A job posting looking for an “AI editor” expects “output of 200 to 250 articles per week.” ChatGPT is being used to generate whole spam sites. Etsy is flooded with “AI-generated junk.” Chatbots cite one another in a misinformation ouroboros. LinkedIn is using AI to stimulate tired users. Snapchat and Instagram hope bots will talk to you when your friends don’t. Redditors are staging blackouts. Stack Overflow mods are on strike. The Internet Archive is fighting off data scrapers, and “AI is tearing Wikipedia apart.”

it’s people who ultimately create the underlying data — whether that’s journalists picking up the phone and checking facts or Reddit users who have had exactly that battery issue with the new DeWalt cordless ratchet and are happy to tell you how they fixed it. By contrast, the information produced by AI language models and chatbots is often incorrect. The tricky thing is that when it’s wrong, it’s wrong in ways that are difficult to spot.

The resulting write-up is basic and predictable. (You can read it here.) It lists five companies, including Columbia, Salomon, and Merrell, along with bullet points that supposedly outline the pros and cons of their products. “Columbia is a well-known and reputable brand for outdoor gear and footwear,” we’re told. “Their waterproof shoes come in various styles” and “their prices are competitive in the market.” You might look at this and think it’s so trite as to be basically useless (and you’d be right), but the information is also subtly wrong.

It’s fluent but not grounded in real-world experience, and so it takes time and expertise to unpick.

Meanwhile, whenever I talk to music people or book people, very quickly the conversation becomes a music industry conversation or a book industry conversation. What matters for music are artists and touring and labels and so on, and what matters for books are writers and publishers and rights and Amazon’s bargaining power in books and so on. These aren’t tech conversations.

Tech needed content to make their devices viable, but having got the content (by any means necessary), and with it of course completely resetting the dynamics of the industry, tech outgrew music and books and moved on to bigger opportunities.

All of this of course takes us to TV, the industry that’s next on the tech industry’s content journey. Just as new technology unlocked massive change in music and (rather less so) in books, it is now about to break apart the bundled, linear channel model of the TV industry (this is especially the case in the USA, which has a hugely over-served pay TV market). As this happens, there are all sorts of questions that follow on: what happens to channels that might be able to make more going direct to consumer (HBO, perhaps); what happens to channels that might benefit from being in a bundle and lose from having to go direct (ESPN, perhaps), where the syndication model goes, and so on, and so on.

Just as for music or books, though, these are all fundamentally TV industry questions. What viewing distribution, what rights structure, what exploitation chain, what relationship between creatives, financiers, aggregators and distributors - these are all southern California questions, not northern California questions. So, what are the northern California questions, and will this end up being any more strategic than books or music?

Amazon and Netflix have entered TV content creation and ownership in ways and on a scale that no-one from tech ever did for music or books. Amazon did try to get into book publishing and has a significant self-publishing arm, but it had little success recruiting existing mainstream authors; neither Apple nor Spotify created a record label.

Netflix, of course, is a TV company, in the context of this conversation - it isn’t using content for leverage for some other platform (Spotify is the same, without the commissioning). But Amazon clearly is using content for platform leverage - as something else to speed up the Prime flywheel. Prime has become a third pillar to Amazon’s business, next to logistics and the ecommerce platform, and Amazon is always looking for ways to add more perceived value to it, preferably with no marginal cost - TV content that it owns outright is exactly that.

You don't close your Facebook account - you just go there less. You might stop paying for the Youtube TV service, but that won’t cut off your access to any other part of Google - nor would anyone want it to - the purpose of these businesses is reach. Nor, really, will you fundamentally change your search behaviour if Google discovers the next Game of Thrones. That is, cancel Prime and you'd lose Amazon, but what do Google & FB have to cancel? Without some platform decision to lock you into, content is marketing, and revenue, but not a lever.

Apple has always preferred a very asset-light approach to things that are outside its core skills. It didn’t create a record label, or an MVNO, and it didn’t create a credit card for Apple Pay - it works with partners on the existing rails as much as possible (even the upcoming Apple Pay P2P service uses a partner bank).

Part of ‘content is king’ was the idea that (at least in theory) content companies can withhold access to their libraries entirely, and in the past one might have presumed that that meant they had the power to kill any new service at birth. In reality, rights-holders have always had too strong a need for short-term revenue to forgo broad distribution, and few of them individually had a strong enough brand to extract a fee that was high enough to justify exclusivity. They always have to take the cheques - individually to meet their bonus targets, and collectively to meet their earnings estimates.

This is a multi-sided market place with too many players on both sides for anyone to exert dominance: Apple dominated purchased music and Amazon dominates ebooks (thanks to the DoJ), but there is no such dominance on the buy or sell side for TV, for now.

The reason Apple TV, Chromecast, FireTV and everything else feel so anti-climactic is that getting onto the TV was a red herring - the device is the phone and the network is the internet. The smartphone is the sun and everything else orbits it. Internet advertising will be bigger than TV advertising this year, and Apple’s revenue is larger than the entire global pay TV industry. This is also why tech companies are even thinking about commissioning their own premium shows today - they are now so big that the budgets involved in buying or creating TV look a lot less daunting than they once did.