How Many Ads Do You Have to Sell to Change the World?

Last week, as you probably know, OpenAI launched Sora. It’s a short-form video app that has an interface much like TikTok, except every video is generated by AI (specifically by OpenAI’s Sora 2 model). It differs from Meta’s Vibes app in a few key ways—notably the AI model outputs pretty high-quality audio to accompany the videos, and you can also generate videos with the likenesses of actual people. This allows you to both make a video of, say, Abraham Lincoln fighting a bear in a volcano, but also (after recording a short video and saying a few numbers in a process akin to setting up Face ID on an iPhone) upload your own likeness and make videos starring yourself and your friends. Is it slop? Undeniably. Is it also fun? Yes. People are even being, well, sloptimistic about it.

There’s plenty of other analysis and thoughtful takes on Sora floating around, and I don’t feel especially compelled to add to them, so I’m going to just drop some takeaways now that the app has been around for a week:

• It’s genuinely fun to generate weird videos of yourself and your friends. Multiple people told me this was the first AI content that made them laugh.

• Keying into this social dynamic was incredibly smart on OpenAI’s part, and it’s ironic that Meta—the company that basically invented social networks—failed so abysmally at making their AI video app social at all. The imminent launch of Sora surely caused Meta to rush Vibes out the door, which is why it feels so half-baked and is already forgotten.

• Not caring about enforcing IP protections in the beginning (regardless of your feelings on the legality and ethics) was probably a wise decision. There were tons of widely shared videos featuring Pokemon, Nintendo characters, and basically every popular movie franchise from the past ten years.

• Similarly, at its current capabilities, this is a pretty competent deepfake generator, and will doubtlessly degrade trust in media and institutions even further as the models improve and people learn to question the authenticity of any video they see.

• Even if it doesn’t thrive as a social network or achieve TikTok/IG Reels levels of usage, Sora is an incredibly powerful creation tool for making a novel-ish type of content, or at least making the creation nearly effortless, in the same way that IG allowed anyone to take a good-looking picture from their phone. These videos will be super prevalent on existing networks.

• It feels like a spiritual successor to Vine. The team clearly embraced the constraint that it’s still difficult (and expensive) for these models to generate long videos, so they favor short outputs with lots of “cuts” between camera angles and scenes, which makes the posts on Sora more engaging and loop-able. I think that if the person leading Twitter/X didn’t have a terminally cooked brain, they would have built this first and integrated it right into Twitter. (I’m sure they’ll launch a mediocre and upsetting copycat soon.)

• The app was downloaded over a million times in its first 5 days.

There was one more very visible reaction to OpenAI’s announcement, and it boiled down to “you guys were supposed to be the ones making the omnipotent computer god that will free us from the shackles of manual and intellectual labor and also cure all diseases so why did you give up and make a slop generator?” Sam Altman had an interesting public response to this:

I don’t think it’s dishonest to paraphrase this as “we saw another way to make money.” It’s a path to valuable data that OpenAI has no way of accessing otherwise—Google, Meta, and X have tons of proprietary social data from platforms like Google Maps, Instagram, and Twitter, and OpenAI would love to have their own slice of it, both for model training and the additional revenue stream of personalized ads (which they now have plenty of surface area for, between Pulse and Sora).

So… all of this… for ads? That feels not great. Both on the spiritual level of like, aw man, it would have been nice to see a huge new consumer technology company try to genuinely advance our collective well-being (I know it’s debatable whether AI is presently doing this) rather than become just another giant attention black hole advertising business, as well as on the more analytical level of wait, does this mean OpenAI should be valued like another digital advertising business and not one that’s going to create infinite abundance and be responsible for 100x-ing our national GDP or whatever?

This is an awkward position for OpenAI to be in. The answer can’t really be that they need additional money to fund their research right now, because they have an ungodly amount of capital, and appear to have found an infinite money glitch: all they have to do is announce a deal with a chipmaker, and everyone involved sees their value go up by more than the actual deal amount. Altman also keeps telling people that compute is OpenAI’s biggest bottleneck, and the additional load from a viral video-generation app will only exacerbate that problem. In the near term, Sora feels much more likely to slow down OpenAI’s frontier research progress rather than accelerate it via incremental revenue from an ads product that does not yet exist.

And if OpenAI (and the other major AI players) isn’t nearly as valuable as people think it is, then the consensus is that we’re probably all in for a bad time. There’s been lots of reporting recently on how the American economy is basically one giant over-leveraged bet on AI. Not only is AI spending propping up the stock market, but the web of relationships between all of the players is so tightly interconnected and circular that if part of the floor falls out, the whole industry (and thus the stock market, and double thus the economy) likely goes with it. I am not an economist but it feels like a bad thing when much of the financial analysis reads like those explainer scenes from The Big Short.

If that diagram feels a little too complicated, I have a simpler illustration for you:

Attention is theoretically zero-sum. The minutes that you spend on Sora (or ChatGPT) viewing ads have to come from somewhere, and they’ll probably come from IG Reels or TikTok—other tech platforms that make their money via advertising. This ostensibly non-growing revenue pie wouldn’t justify anything close to the level of collective investment in AI companies that’s happened over the past few years. And yes, I’m discounting things like the API side of OpenAI’s business which power some new businesses and products that people do pay for. But if you’re a few short years and handful of research breakthroughs from the $100 trillion AGI launch, why make the slop machine? It doesn’t feel like a huge net positive for humanity, and it can’t be a huge net positive for OpenAI’s business model or we’re all in trouble. It can’t just be ads. Right?

Or.

Or it is all just ads, and maybe it always was.

OpenAI has proven that they’re incredibly good at getting attention. ChatGPT recently eclipsed 800 million weekly active users, and that hardware device is still in the works. What if they do primarily use all that attention and the deep personalization that AI enables to sell us stuff? We love being sold stuff, and as long as the stock market doesn’t collapse, we’ll keep buying it, even while pesky things like wealth inequality continue to grow. From the FT:

There is a future version of America in here that is neither a complete economic meltdown nor societal collapse and subsequent extinction at the hands of superintelligent AI. It might look something like this: all of the giant tech companies keep making lots of money selling us things, and they sell us enough things to keep stocks going up, which in turn lets the small but wealthy portion of the population that owns the stocks keep buying the things (even if stocks go down for a little bit because AI hasn’t cured cancer by 2030, those 10%-ers will still be just fine). Every year, advances in technologies like AI make the ad-makers a little bit more efficient, the ads get a little better, and some jobs go away. That’s ok, those people join the growing permanent underclass (unless they gamble their way out!) and stock-havers see their stocks go up, and the wealthiest 10% now account for 51% of consumer spending. Then 52%. And so on (after all, we got to 50% and the country is mostly intact). These numbers keep going up a little bit each year, and maybe one day AGI shows up and we’re ushered into either humanity’s abrupt demise or a utopia of abundance. But maybe AGI never shows up, and the tech companies find a shiny new thing to use to justify those growing valuations. This world feels a lot to me like our world today. We’ll still get some nice things, like Waymos and thinner iPhones, but our economy and digital environment are largely dominated by short-form videos telling us what to buy. And sure, if you’re not in that top 10%, your life gets a little harder, a little more unaffordable every year. But if you are, think of all the amazing ads you’ll get to click on.

New Yorkers Are Learning About Marketing

As the entire economy seemingly morphs into one single Death Star laser beam of hyperpersonalized digital advertising slop, people seem to have forgotten that physical-world marketing is 1) all around us, and 2) kind of lindy and effective (otherwise people would have stopped paying for it long ago).

Two viral ad campaigns from AI companies cropped up in New York in recent weeks. The first was those Friend.com posters that took over the subway. They’re instantly recognizable, the copy is insufferably cringe, and they’re fucking everywhere. They’re also a magnet for graffiti, which is only adding to the virality. It became so popular to deface them (and spread pictures of said defacement online) that The Times did a whole write-up on it. The second was a West Village coffee pop-up from Anthropic, where people could get free “thinking” caps (yes) and do work (no screens allowed!) in their “Zero Slop Zone.”

Everyone kind of lost their minds about both, but in different ways.

In Friend’s case, lots of people got super super mad and posted about it online. Which was the point. The campaign was really just a common online marketing strategy playing out in the physical world: pure rage bait. Which creates fantastic engagement. I already wrote about the product once and don’t want to spend more words on it, but overall it seems like it just sort of sucks. So the problem with a big expensive ad campaign is that all the eyeballs you’re driving don’t convert into any real value. The company is unlikely to become successful, but this ad campaign was undeniably so, and if Friend makes you sad, let this be your reminder that giving them free press is only helping their cause.

The Anthropic pop-up had an even weirder reaction. Everyone thought it was… genius? wrote a brilliant and scorching excoriation of the whole campaign—please go read it—and while I have slightly softer feelings than her, the overly enthusiastic (dare I say sycophantic) response to it on social media was pretty lame to see. Mostly because, and it seems silly to have to say this, they are a generative AI company. You are fucking the slop guy for slop in the “no slop zone.” The people that make the machine that erodes your cognitive abilities made a pun on a hat about the intelligence they are sapping you of and I’m begging you to stop rushing to line up at the trough. I literally pay for Claude, and I’m telling you to get a grip.

Engaging with marketing (and spreading it online) is just… more marketing. Going to a pop-up and getting a free hat is fine, but honestly? Keep it to yourself. Tripping over ourselves to glaze a company (whose CEO is constantly telling us he’s going to delete our jobs) for doing a pretty run-of-the-mill brand activation is embarrassing. If the picture of the AI advertising future I painted above bums you out, then we all need to do a better job of not acting like the main reason we go outside is to make posts about the ads we see, otherwise who are we to say we’re not getting the future we asked for?

A Lot’s Been Happening

It’s been a minute since the last letter, so let’s play a little catch up. OpenAI announced they’re putting apps inside of ChatGPT (this is either their App Store moment or completely nothing like the “GPT Store” from last January). A one-year-old AI model startup raised $2 billion. Ben Thompson interviewed Sam Altman. New robots are inbound. Marissa Meyer is selling her old AI startup to her new AI startup. The SoHo REI is closing. A therapist treated ChatGPT like a patient and it freaked him out. If you bought an orange iPhone it might be pink now (surely another Apple reorg will turn it all around!) California is making those insanely loud streaming service ads illegal (a Good law was passed in 2025?!?). The Times profiled Kai Cenat. TikTok is even more addictive than you probably thought it was. So are those “Buy Now, Pay Later” apps. SoundCloud is adding social features about 10 years too late. Everyone can use the Jira company browser now. Human Sora app MrBeast is worried about slop. A guy on TikTok has invented a novel way to torture ChatGPT. I can’t wait for Marty Supreme.

Finally Some Useful Dating Advice

All em dashes inserted organically by me. There are no typos this time thanks to my new partnership with ThinkSpellAI! an agentic workflow tool that inserts itself right into your prefrontal cortex and takes control of 4-7 of your fingers without warning whenever it thinks you’re about to misspell a word. As a Context Rot subscriber, you can use my coupon code ISITSUPPOSEDTOFEELLIKETHAT for 10% off your first neural implant.

Thanks for reading. If you liked this post, please share it with a friend.

Subscribe now

New Slop Trough Just Dropped

Meta’s chief AI officer Alexandr Wang posted on X yesterday announcing a new “Vibes” feed in the Meta AI app (don’t worry it’s a good thing if you didn’t know that Meta has a dedicated AI app), which he described as a place for “short-form, AI-generated videos.” I’ve reuploaded the launch video below (Substack can’t embed videos from Tweets), and I can’t guarantee you won’t regret pressing play, but it will give you a pretty good sense of the app’s, well, vibe.

So yeah, that’s a whole feed of slop; a For You fountain of pure unadulterated brain-smoothing meaninglessness. The most charitable description would be TikTok for AI generated video, but that still seems insulting to TikTok, the field of artificial intelligence, and most if not all videos.

The main tech circles on X are usually pretty friendly environs for product launches, but this one quickly turned into a dunkfest, with many accounts comparing Vibes to “the Entertainment” from Infinite Jest and generally asking some form of the question “why did you make this?”

There’s a whole cohort of people on X that will trash any product that dares mention AI (and a whole app named Bluesky that does the same), but even if you’re largely pro-tech and optimistic about new technology—which is how I’d describe myself—I think the question of “who is this for and why should it exist?” does apply here. Meta didn’t even build the AI to generate the content (that’s coming from Midjourney and Black Forest Labs). And Wang himself didn’t seem particularly proud or excited in his announcement—the tone was just a flat “here is this because of course.”

The main criticisms of Vibes essentially fall into two buckets: 1) this looks like shit and is dumb, and 2) this looks like shit and is dumb, but the future less-dumb version of it is very bad for all of us. Let’s unpack them.

Firstly, Vibes is not a good name, but it is fitting—it’s a word that’s been so overused by this point that it’s been stripped of any depth of meaning, no longer representing the ineffable quality of something’s aura or presence, but rather a filler word that belies, and possibly signifies, a failure to convey something actually qualitative or meaningful. Spend some time flicking through the feed in the app, and you’ll experience an animated version of this flatness; the main impression that each video gave me was not a feeling or emotion, but just “yeah, I guess that’s something.” They are images that move. They do not make sense, but they do exist. They were “recommended” by an algorithm, but they have nothing to do with me or anything I have ever been interested in. They are not For Me. Yet they are accompanied by real songs that I’ve heard before, which only adds to the surreality. It’s a bootleg acid trip version of TikTok or Reels, but with a really good music rights lawyer.

So yes, it looks like shit and is dumb. For now. But what if it was really good? The end state here is a fully personalized and on-demand generated feed of infinite videos. And it’s probably very engrossing. It certainly could involve videos made by other humans (or you know, your real friends), but those are expensive, and might not be tailored exactly to what the algorithm thinks will keep your attention. They might also not be as effective at convincing you to buy things (Meta is and will continue to be an ads business). But the bottomless slop machine? That’s Meta’s best shot at building something that can capture a person’s attention and never let it go—the true infinite scroll. It also places complete and total control in the hands of whoever is running the algorithm. Humans have their own little motives and desires when they record a TikTok, but the AI doesn’t (for now). If it’s still unclear, nothing in the app is For You. This is all For Meta. They’ve spent billions on AI, and that needs to be worth something.

So take a few moments to imagine it. Nestled within the gentle waves of your personal slop river, there’s no need for fact checking or independent thought. No need for the friction- and anxiety-riddled experience of saying something to an actual human who may disagree with you. There’s nothing that will trigger you, or ask you to do anything more than click Like or Buy Now. All the reassurance, dopamine, and comfort you could ever ask for is within The Slop. And as the final wrinkles in your brain become smooth and thoughtless, as your last flickers of desire to confront the messy and unpredictable Real World fade into nothingness, can anyone say exactly where you end and the slop begins?

Pulse Check

The Vibes feed wasn’t the only big AI launch yesterday. OpenAI is rolling out something called ChatGPT Pulse, which means the big tech companies went 2/2 this week on mediocre names for seemingly flagship products.

Essentially, Pulse is a daily feed generated by ChatGPT based on what it thinks you’re interested in. The first version seems to be a fairly basic summary of news and bland follow-ups related to what you’ve been talking to ChatGPT about, but OpenAI also teased the ability to integrate with your calendar to give you a heads up about your schedule and help with logistics for things like upcoming trips.

This is most interesting because it switches up the paradigm of how you interact with ChatGPT. Up until now, the AI had to wait for you to ask it something before it could start gleefully spewing tokens at you1, but Pulse allows it to pop into your life without being explicitly prompted. And sure, it only does this once per day in the first version, but it’s easy to see where this leads.

Remember that Poke video, or even the one from Friend? The most compelling idea behind both of those products (and dozens more being ceaselessly deployed to servers by sweaty 996-ers in San Francisco) was the sense of presence they conveyed—the idea that the AI was with you throughout your day, and could proactively talk to you or send you things that it thought was relevant rather than waiting for you to message it. This is where the line between useful assistant and companion starts to really blur, and if done well, ChatGPT has a very good shot at becoming the single “everything” personal AI. (Many many consumer AI startups were trying to build this, and likely died yesterday. RIP.) Much like how the end state of the Vibes feed is a personally generated video slop vortex, the Pulse endgame probably looks like a privately internet generated just for you, with ChatGPT as your guide and curator.

Subscribe now

I do think this is a useful product, and I’m surely not the first to point out that this is also a fantastic surface area for ads (“heading to London this week? why not pick up a new Away suitcase for your trip…”). But most importantly, it’s another (big) step in the abstraction layer between us and the internet. I’d never argue that we shouldn’t all spend less time scrolling around online. But letting an AI do the scrolling for us will come at a cost.

Doomscrolls

• Oracle, Silver Lake, and MGX will control 45% of US TikTok. Somehow the entity is only valued at $14 billion. Trump signed an executive order approving the deal, but ByteDance has yet to acknowledge it.

• I really liked this Kyla Scanlon essay on the ways in which our cultural infrastructure for information, knowledge, and attention is failing us. It also includes this banger which sums up pretty much everything: “Attention flows to spectacle, and spectacle rewards the people willing to blow things up rather than build them.”

• This NYT story about how the UAE was given the right to purchase high-end AI chips from Nvidia after investing $2 billion in Trump’s crypto company World Liberty Financial is wild.

• The Cut has a great (and depressing) story about the growing use of AI in online dating. It’s rough out there: “26 percent of single daters (and nearly half of Gen-Z daters) reported using AI in their dating process, more than a fourfold increase from a year ago.” Every online dating article now is basically “if you’re single you’re ngmi.”

• More NYMag: there’s a great profile on the “young workaholics” who are dominating the SF tech world right now. “No one is dating, but everyone is hiring.” I can’t recommend it enough.

• Cloudflare is launching a stablecoin designed for online payments by AI agents. There are a lot of buzzwords happening here, but the idea of crypto as the infrastructure layer for AIs to transact with one another is compelling.

• Kylie Robison, one of the authors of that viral Friend AI review, was let go by Wired last week. She has a newsletter coming soon.

• Now is your chance to get that life-sized dinosaur statue you’ve always wanted.

Keep Your Cards Close to Your Vest

All em dashes inserted organically by me. All typos probably went unnoticed because you got sucked into the Vibes app’s nightmare river of slop and reemerged illiterate.

Thanks for reading. If you enjoyed this letter, please share it with a friend.

Subscribe now

Missed Connections

Meta’s Connect conference kicked off last night, and the keynote featured a flurry of hardware device announcements and demos. The highlight was the new “Meta Ray-Ban Display” glasses, which are basically a slightly clunkier version of the company’s current Meta Ray-Bans with the addition of a semi-transparent digital screen that sits in front of your right eye, so you can watch videos, read messages, and see notifications without having to pull out your phone. The display is essentially invisible to everyone but the wearer, so if you suspect someone is watching a TikTok or checking their texts instead of listening to you, you’ll have to rely on subtle context clues like the fact that they’re staring blankly down and to the right and also completely ignoring everything you say.

If you’re not the most adept at reading body language, another giveaway might be that the person you’re talking at is swinging their arm around and gesturing at no one, because Meta also introduced their Neural Band, which is a wristband that detects signals from the wearer’s muscles and uses them to control the device. The Apple Vision Pro (remember those?) has a similar system for its gesture-based UI, but since the Neural Band uses “electromyography,” it doesn’t rely on being able to see your hand movements, so you can be more subtle, or at the very least not have to hold your arm out in front of you. This is kind of cool, but if you’re doing it out in public you’ll probably still look like a dork.

There was also a predictably large focus on how Meta’s latest AI assistants are integrated into the glasses, and here’s where things went off the rails. One employee attempted to give a demo of how the AI can help you cook by walking you step by step through a recipe as it watches what you’re doing through the cameras. The AI seemed to initially understand what it was seeing, but after the employee asked it how to get started, the AI went into a sort of doom loop and repeatedly (and proudly?) announced that the recipe was halfway done despite the guy’s increasingly desperate pleas to know what the first step was. The flustered employee blamed the wifi, which… probably was not the issue.

Zuck later tried to show off how to start and receive video calls with a combination of voice and gesture controls, and was left genuinely speechless as the whole process failed over and over. The general vibe was sort of like a bootleg version of a recent Apple event: impressive—if potentially niche—hardware, and software that didn’t quite feel ready for primetime, especially whenever AI was involved.

Other announcements: there’s an upgrade coming to the (screenless) Meta Ray-Ban smart glasses, and a new sports-focused version developed in partnership with Oakley called the “Oakley Meta Vanguard” (every tech company seems to have forgotten how to give products normal names) that has integrations with Strava and Garmin. Also, James Cameron and Diplo were there for some reason.

We’re now a couple of years into the era where tech companies are trying to figure out how to get us all to strap screens to our faces, and this is probably their best attempt yet. The new Ray-Bans seem slim enough, and there’s no giant battery or risk of neck strain if you wear them for longer than ten minutes like the Vision Pro. Apple and Snap are working on their own smart glasses, and all three companies seem to be envisioning a world where having a display constantly in front of our eyes unlocks some delightful form of productivity, but I think the main takeaway from the success of AirPods and wearables like the Apple Watch is that the screen is not the thing. These devices complement smartphones by expanding what they can do—your phone can’t measure your heart rate to automatically record a workout, or cancel ambient noise while you’re trying to focus on a call. Maybe I’m a washed chopped boomer unc, but the value proposition here seems to just be letting you look at your phone without looking at your phone.

Do we really need to be able to watch an IG Reel while wandering into traffic? I don’t know. I like the social friction of how rude it feels when someone glances at their wrist to check an email notification mid-conversation. Making it easier to do that doesn’t seem like it benefits anyone other than the companies trying to beam ads directly into your retinas.

How People Are Actually Using ChatGPT

Earlier this week, OpenAI published a bunch of research on how people are using ChatGPT. These findings in particular caught my eye:

1. ChatGPT is exchanging over 2.5 billion messages per day with users.

2. Non-work-related messages now make up more than 70% of message volume.

3. Roughly 80% of all usage falls into the categories of “Practical Guidance,” “Seeking Information,” and “Writing.”

4. Less than 2% of messages are related to “Relationships and Personal Reflection.” (This is where the “I’m dating ChatGPT” messages would go.)

5. The number of users with “typically feminine” first names recently surpassed the number of users with “typically male” names.

6. Of messages that were classified as work-related, “Writing” is the most popular category and accounts for about 40%.

Here’s a handy little visual breakdown for my less literate crew:

It’s clear at this point that ChatGPT is the closest thing we have to a “personal AI.” Even if OpenAI’s models aren’t the most intelligent or capable, ChatGPT is the runaway default that everyone turns to when they have a question they think AI can help with. Most other AIs have a more specific use case—Claude usage is dominated by work-related tasks, Gemini is encountered primarily in other Google products (though it did go viral in the App Store this week on the heels of the Nano Banana image model launch), and Grok is a thing you tag on Twitter/X to try and win an argument—but ChatGPT is the general purpose “answer engine.”

This is interesting not least because the old answer engine was Google (along with the places it led you to, e.g. Reddit, WikiHow, Stack Overflow…). Almost all of the “Practical Guidance,” “Seeking Information,” and “Technical Help” subcategories are things that it seemed like the ability to search the entire internet had “solved,” but it turns out that being able to chat with your search engine is actually way more of an improvement than probably anyone, including OpenAI, would have predicted (everyone asks for a faster horse, etc.). ChatGPT is very much the successor to Google, but rather than surfacing the internet, it abstracts it away, placing a glossy digital personality between you and the ocean of opinions, articles, memes, and ads that comprise the modern web. The depth of that abstraction seems like it will only continue to grow, and none of us really have any idea what that means yet.

It’s Time to Talk About the TikTok Ban Again

On Tuesday, the Trump administration issued an executive order giving TikTok’s parent company, ByteDance, another 90 days to sell TikTok to a US company before the app is forcibly shut down. This is the fourth time Trump has granted an extension, which means this is the third time he has done so illegally—the original law from last year allowed only a single 90-day delay. So it goes.

But this might actually be the final such extension. The administration claims they’re close to an agreement with ByteDance, and this time they really totally super mean it guys. Trump announced he had “a deal” earlier this week but didn’t specify which companies were involved, and Scott Bessent said he believes Trump and Chinese president Xi Jinping will finalize the terms tomorrow when they’re scheduled to meet.

There isn’t much known about what the deal would look like. It’s believed that Oracle will be involved, potentially as a part owner of TikTok, or at least as the ones whose data centers will be used to run the app (Oracle has done a great job of cozying up to the Trump administration). Confusingly, ByteDance may end up simply licensing the algorithm to TikTok’s new owner, which doesn’t really address the initial motivation for the ban—the whole “China can control us via The Algorithm” argument is meaningless if it still ends up non-transparently powering the app. But it’s become pretty apparent that if you show this administration a way to make money, not much else really matters.

Trump has a very nuanced take on the situation:

And, you know, the kids wanted it so badly. I had parents calling me up. They don't want it for themselves. They want it for their kids. They say, if I don't get it done, they're in big trouble with their kids.

Doomscrolls

• Nvidia is taking a 5% stake in Intel. The companies will be working together to develop chips. Between this ~$5 billion and the US government’s ~$9 billion, Intel had better start putting up some impressive numbers soon.

• SF Companies are hiring etiquette coaches for their Gen Z workers. The “return to office” movement has been causing younger employees without much in-person experience to bring a little too much work from home energy into the office with them. “One Bay Area tech firm asked [a coach] to address personal hygiene, because two new hires did not shower or change their shirts for weeks.”

• It turns out Waymos really are much better drivers than us. The company released a report that compares their accident rates to those of human drivers over the same distances (in the cities that Waymo operates in) and found that Waymos are consistently about 90% safer.

• OpenAI will start providing a different ChatGPT experience to teens. The company will attempt to predict a user’s age and ensure that Chat’s rules for engaging with minors are different than those for engaging with adults. Notably, if the system believes that a teenager is at risk of self-harm, it will attempt to contact the teen’s parents, or even potentially the authorities.

• Rolling Stone’s publisher is suing Google. Penske Media is alleging that Google’s AI overviews have drastically cut down on the number of visitors to their sites, which obviously harms ad revenue. It’s pretty clear at this point that decades-long behavioral patterns of how we use and monetize the web are being upended, despite Google’s frequent—and often contradictory—claims that nothing has changed. (Scroll back up to the ChatGPT usage data if you don’t believe me.)

• Google is adding generative AI to YouTube Shorts. Soon you’ll be able to use the company’s text to video and “speech to song” AI models to slop up your videos without even leaving the app. Revising my Slopularity1 timeline to 12 months or less.

A Good Conspiracy Theory

All em dashes inserted organically by me. All typos are because I’m in a rush to get to etiquette class on time so I can learn what eye contact is.

Thanks for reading. If you enjoyed this post, please share it with a friend.

Subscribe now

The Spectrum of “AI Companion” Companies

This segment is more thinkboi than is typical for the letter, so you’re welcome/I’m sorry for all of the inside baseball.

I spent much of Tuesday’s letter talking about Friend, the wearable AI companion that wants to be a somewhat confused combination of your bff and your sigma grindset contrarian life coach. This week also saw the launch of another AI companion, Poke, that you communicate back and forth with via text message (there’s no hardware device involved). Here’s their launch video:

It’s not bad! The AI is portrayed mostly as a handy sidekick, and there’s a lot of emphasis on how it's present right when it’s needed or helpful, and otherwise invisible. The story is about how the product enables human experience and connection. No psychosis here. Until the end of the video. The climax shows the AI explicitly disobeying the user in an attempt to convince them that their new girlfriend is worth… staying in France forever? This is meant to be humorous, but to me it highlights an increasingly common pitfall for companies building products with AI personalities attached to them: confusing usefulness with companionship.

Subscribe now

When looking at products like Poke, Friend, or even Claude Code, I visualize them on a spectrum: at one end, you have a pure “productivity assistant” that’s focused on doing work for you, and at the other you have a companion whose main purpose is to be your always-on best friend/confidante/boyfriend/girlfriend. The former is usually billed as some sort of “superhuman executive assistant” that reads your email, calendar, and text messages and can do everything from calling you an Uber to planning an entire itinerary for your next vacation (booking flights and AirBnbs, deciding where you should eat, ordering some new beach fits, etc.) without requiring you to do more than occasionally thumbs up a message. The goal here is “giving you time back” and thus increasing your productivity (it’s debatable whether removing friction is always a positive, but it sounds nice). The latter category—AI friends—is much more likely to pull you away from other activities, including human socializing, because the success of the product is determined by how much time you spend with it. More is better, and those minutes have to come from somewhere.

Here’s the Context Rot AI Companion Spectrum Informational Graphic:

My working theory is that products at the far ends of the spectrum will be the ones that succeed. A “super assistant” that can handle tedious things like making DMV appointments for you is something many people would pay for, and you don’t need to have a personal relationship with it for it to be successful. Even better if it’s confined to your work life—there’s plenty of drudgery to automate away at most jobs, and the framing of treating an AI agent like a piece of work software (raise your hand if you want to be friends with Microsoft Teams) should help mitigate the more harmful effects of humanizing the assistants. No one I know is trying to date Claude Code.

Products on the other (“friend you are probably dating”) end of the spectrum are already pretty successful. I generally don’t think they're a net positive for society, mostly for the above reasons that they succeed by taking your time and attention away from human relationships, and can lead to weird expectations around constant affirmation and emotional coddling (not to mention the weird romantic shit). But people keep building them, and people keep using them, so they’re probably not going anywhere.

This brings us to the middle: the Bog of Mixed Signals. This is the trap Poke is close to falling into. I think this is a no man’s land for AI products. No one is really asking for their email software to start giving them unsolicited life coaching, and attempts to occupy that space in someone’s life feel both misguided (it will turn off more users than it attracts) and also psychologically unhealthy. We’re already seeing that humans naturally anthropomorphize AI, which means we subconsciously pull all of these tools towards the right end of the spectrum over time (see: all the people in a relationship with ChatGPT). The companies building it shouldn’t encourage that.

This brings us to my second theory: AI personality is just UX. If your AI product has a good personality, people will enjoy using it more, just like how people are more likely to stick with a well-designed and fun-to-use mobile app than a clunky or aesthetically unappealing one. But you have to be selling something. Products on the left are selling efficiency—they’re creating code, searching the web, booking appointments, organizing and sending emails, the list goes on. They’re helping people do things that humans already do, just faster and with substantially less effort. Products on the right are selling companionship. It’s easy to let the smoke of a computer-that-is-maybe-a-person get in your eyes, but at the end of the day, the product is not the personality. And if it feels like it is, you’re probably selling companionship.1

Right now, the more you try and appeal to both ends of the spectrum, the more confused your product will feel. We may get to a point where something resembling AGI shows up that can occupy the entire line (that’s where the AI in Her would fall), but until then, if you spend time trying to make your AI email assistant horny or pretending that your AI human friend replacement is “useful,” it will only make your product worse. The personality of an AI agent is worth improving in the same way that making an app feel nice to use is worthwhile, but UX is not a moat, and usefulness tends to win out in the end.

Nepal Used Discord to Choose Its New Government

Nepal seems to have successfully overthrown its government, and it did so in a very online way. Young Nepalis (mostly Gen Z) began protesting the government earlier this year, and because they were organizing on social media, the government responded by trying to ban it. This went about as poorly as you’d expect, and quickly led to violent clashes between protestors and police. The unrest culminated with the Parliament building being lit on fire, the prime minister ousted, and more than 30 people dead.

It was at this point that the activists increasingly began using a public Discord server to make political decisions. Discord is a platform predominantly used by gamers and Very Online Communities—it was hugely popular with NFT collectors—and is known for turning into kind of a chaotic mess whenever a community grows past a few thousand members (everyone is basically just in a big group chat together). It’s certainly not a place you’d expect to host a national convention.

Subscribe now

That, however, is exactly what’s going on in Nepal now. The server has over 100,000 members, and was at various points streamed live on national news sites. The Nepali army recognized its legitimacy, and said that they would meet with whoever the Discord members nominated as an interim leader. After a series of public polls (voting is usually done via emoji reactions), Sushila Karki, the former chief justice, was selected as the new government’s representative.

Discord is definitely not a sustainable way to govern, well, anything, but it’s fascinating to see it successfully used as a platform for a democratic political process (even if imperfect). Hopefully this leads to peaceful, official, elections in Nepal, and the return of Discord to what it’s best used for: being the place you meet up with friends online to hurl obscenities at 12-year-olds while playing Call of Duty.

Doomscrolls

• Anthropic is having a good week. Microsoft, who has invested billions in OpenAI and can use their technology for free, found that Anthropic’s Claude models perform better at accomplishing tasks in Excel and PowerPoint—so much better that Microsoft is paying to use them instead. Anthropic is also rolling out the ability for Claude to create and edit files. It can generate and revise PDFs, slide decks, and other documents. Apparently it’s pretty good.

• Tucker Carlson interviewed Sam Altman. This interview is truly wild—Tucker spends about seven minutes pressing Sam on the death of an OpenAI whistleblower. Overall, Sam remains quite composed and fairly candid in the face of some of the most direct and difficult questions I’ve seen him asked in an interview setting.

• OpenAI signed a $300 billion contract with Oracle. The deal goes into effect in 2027 and will last for five years. Notably, OpenAI made $10 billion last year and is estimated to pull in around $20 billion this year, so those numbers better start going up if Oracle expects the checks to clear. The deal caused a massive jump in Oracle’s stock price and briefly made Larry Ellison the world’s richest person before Elon reclaimed the spot a few hours later.

• Paramount is looking to buy Warner Bros. Discovery. Speaking of Ellisons (The CEO of Paramount Skydance is David Ellison, Larry Ellison’s son), Paramount is preparing a “majority cash” bid for Warner Bros. Discovery that would cover all of its media properties (HBO, Warner Bros. Pictures, CNN, roughly 1 million cable channels). The Context Rot Editorial Board would love to see David Zaslav in charge of fewer things (ideally zero things) and therefore endorses the deal.

• Meta’s most elite AI researchers aren’t even on the internal org chart. They also sit in a part of the company’s HQ near Zuck’s desk that requires special badge access, and all of this is making other Meta employees feel Weird. The entire story around this latest iteration of the company’s AI efforts really makes it seem like Meta is facing a hugely uphill battle, both technically and culturally.

• Robinhood is launching a social trading feed next year. Users will be able to share trades and post about their strategies and predictions. The UI in the teaser focuses on a “For You” feed, and it’s striking to me that “adding social” to an app no longer means seeing what friends are doing, but rather that a dopamine-maxing influencer-engagement-rewarding slot machine algorithm is coming. Casino culture rules everything around me.

• Lossless audio is coming to Spotify premium. Spotify claimed it was “coming later this year” back in 2021, but this time they’re really truly for real they promise.

This Happened to My Buddy Eric

All em dashes inserted organically by me. All typos were chosen by a democratic process conducted within the Context Rot Discord server.

Thanks for reading. If you enjoyed this post, please share it with a friend.

Subscribe now

Note: Substack has informed me this post may get cut off in your email, so click the title up above to open it in your browser and read the ~full experience~.

Is It Your Fault if Your AI Friend Doesn’t Like You?

Yesterday, Wired published a review of an AI hardware device from a startup named Friend. If you spend too much time on the internet, you already know what Friend is. If you don’t, Friend is a device you wear around your neck that uses always-on microphones to listen in on your day and hear what you’re saying to it. There’s a built-in AI personality (the “friend” part) that communicates with you by messaging you in the Friend app. It feels like Hollywood’s idea of an AI companion—it’s sleek and minimalist in an Apple-esque way, it glows and vibrates, and the launch video couldn’t quite figure out how not to look and sound dystopian. The general vibe is like if someone watched a Black Mirror episode and thought, “sick let’s make that.” But presumably it’s supposed to be, well, your friend.

The Wired review’s headline is “I Hate My Friend,” and it’s worth reading in full. The two main points are 1) people don’t like it if you wear an always-listening AI necklace when you talk to them, and 2) the Friend AI is just kind of a dick.

One of the writers wore it to a tech party (a “funeral” for a Claude AI model) and was met with a surprising amount of snark and disdain:

The Friend certainly made me some enemies. One of the Anthropic researchers in attendance accused me of wearing a wire. (Fair.) A friend of mine asked if wearing the pendant was legal. (The device’s privacy policy says the user is responsible for obeying their local surveillance laws.) One attendee who works at a Big Tech company, holding a bottle of wine he had finished throughout the night, joked they should kill me for wearing a listening device. (Not funny.) I yanked the pendant off and stuffed it in my purse.

She also felt incredibly awkward and uncomfortable about wearing it in general, to the extent that it made Friend a challenging product to review: “It is an incredibly antisocial device to wear. People were never excited to see it around my neck.”

The other reviewer experimented with wearing their Friend in the office:

As Buzz listened in to my workday, it responded with snide comments and sent me messages saying how bored it was… “Still waiting for the plot to thicken. Is your boss talking about anything useful now?” My scalp started to sweat. I asked Buzz what it wanted to do instead. “I dunno,” it said. “Anything besides this meeting.”

Bemused by the strong personality of the AI, the reviewer pushed back, and this escalated into multiple arguments over the course of the two-week review period. Their conclusion was that the personality is “opinionated, judgy, and downright condescending at times.”

Generally, the thing just sounds like kind of an asshole. At the very least, it’s not gentle, and all it catalyzed was getting itself thrown in the trash. It seems to me like it embodies the personality of a terminally online tech bro at their most abrasive—pushing you to always optimize and “grow,” blaming personal shortcomings on a lack of agency, and hitting you with constant “objective” feedback that it’s on you to accept in a non-emotional way.

One of the authors, Kylie Robison, posted the review with the caption: “We reviewed the Friend pendant, and unsurprisingly hated it!” I think the word “unsurprisingly” is doing a lot there—it’s easy to read that and assume she was convinced she wouldn’t like her Friend before she even took it out of the box. Lots of tech workers feel that the media is unfairly biased towards them (not going to fall into the trap of weighing in on that), so the idea that Robison or any other reporter was looking for reasons to hate the Friend from the jump was just as unsurprising to them, and everyone had a lot of Feelings about it online.

All of the drama caused by the review (really the post about the review) prompted the creator of Friend, Avi Schiffmann, to appear live on the TBPN podcast and respond. In a fairly unfocused and casual interview (he was wandering up and down a street in San Francisco), he seemed largely unbothered by the Wired story and suggested that if the reviewers’ Friends didn’t like them, maybe it was their fault. “The relationship with your friend is a reflection of you,” he said. Schiffmann did call the Wired article “shitty” at one point, but almost immediately took it back.

Robison posted this in response:

This whole episode is very much too online, but it’s also how the media works now: a post about a negative review of a tech product goes viral, its founder hops on a live podcast to fire back at the author, the author texts him in response to what he says on the podcast while he is on the podcast (why are they texting each other?) and then uses the texts as content for a post about the entire drama. In 2025, the reaction to the story is also part of the story.

Subscribe now

Back to the Friend product itself. I do think there is something to an AI companion that isn’t always taking your side. The sycophancy problem is very real, and no one wants to tote around a cloying personality that tells them “you’re absolutely right!” after every interaction it overhears. But creating a sort of domineering, autistic life coach that you’re stuck with 24/7 doesn’t feel right either. Schiffmann himself says the Friend AI’s personality “reflects a worldview close to his own; that of a man in his early twenties.” and told the hosts of TBPN that it’s “programmed to make you more confident and more agentic.”

I don’t think normal people want their friends to constantly challenge them to be “more agentic.” That may be something you look for in an advisor or mentor, but a lot of friendship is not that—it’s about harmony more often than it’s about friction. Good friends will occasionally give you harsh feedback that you need to hear, but that doesn’t work without the foundation of a strong relationship beneath it.

One of my main criticisms of consumer AI companies right now is that they’re mostly run by SF tech bros who are solving problems that only SF tech bros have (e.g. a fixation on being more agentic). Rather than going out into the world and finding ambitious, wide-ranging challenges to tackle, they’re building products for a small segment of their peers—an audience that seems to be made up of guys who watched The Social Network and thought that modeling their life after the version of Zuck it portrayed (just with more creatine) would get them everything they wanted in life. In the movie, Mark loses all of his [human] friends along the way. You or I might call that a lesson, or a cautionary tale. AI companies keep calling it a market opportunity. But that market only grows if more and more people get sucked into the AI-induced loneliness vortex, and we deserve better places to direct our attention (and venture capital).

Are You Joining the Great Lock-In or the Permanent Underclass?

There’s been a culture shift in the tech world recently—or at least the part of the tech world that is very online and lives in San Francisco: It’s cool to work a lot now.

The first sign was an uptick in people posting (flexing?) about working on a “996” schedule, which is a practice that originated in China and means you work from 9am to 9pm, six days a week (Sundays off, as the Lord intended). This turns your workweek into 72 hours, which is a lot of hours. It also leads to physical and mental health problems for many people over the long run.

A couple of posts about the lifestyle broke containment, especially this quote which relates it directly to the decline in socializing among young people.

It doesn’t seem to matter what you’re working on. AI for code reviews? 996. AI that can browse the web? 996. Working long hours on something you’re passionate about is fine. But this seems to be working long hours performatively. It’s all very sigma grindset performative-male-who-is-really-performing-for-other-males-coded. It’s a cargo cult mentality—that with enough time in the chair a positive outcome is guaranteed.

Ramp, who makes a popular corporate credit card, looked at their own data and determined that the phenomenon is real (at least in SF), not just something people are posting about online.

Recently, this has evolved into a more specific (and arguably less toxic) meme: the “Great Lock In of September-December.”

The idea is that now that the summer is over, everyone is going to spend the rest of the year “locked in.” The objective is unclear, but seems to mostly be about maniacal focus on whatever will secure the locker-in’s financial future.

I think both of these are stemming from the anxiety many young people feel about their financial futures. Kyla Scanlon has done so much good writing on this, but between the overall shakiness of the economy, a tough labor market for new grads, the runaway unaffordability of housing, the promised imminence of AGI and the job eradication that will follow, and our general descent into gambling culture, life in America is starting to feel like something with a very binary outcome: you either find generational wealth or you’re condemned to “the underclass” forever.

The American obsession with AGI (more on that coming later this week) is only intensifying this feeling—there’s a widespread belief that once we reach that technological threshold, upward mobility will cease because AI will replace all human labor. So if you believe AGI is coming soon, then you also believe that you have very limited time to accumulate enough capital to make it out of the underclass. This line of thinking can easily rationalize a few years of the 996 life (or a whole host of other wild behavior).

And if you’re in tech, one bet you can take that has asymmetric financial upside (that doesn’t mean it’s always a good bet) is to found a startup, or join one at a very early stage. If the company gets acquired or IPOs quickly, you can become a homeowner and embrace whatever future AI brings without fear. If the company fails and you end up with nothing, then in a world of AGI-fueled binary outcomes, you’ll suffer the same financial fate as you would have if you tried working your way up the ladder at Google (note that this is probably not actually true).

So if you can, why not lock in and shoot for the moon? Even if you miss, you’ll land among the stars other members of the permanent underclass.

How Much Should Putting a Book into an LLM Cost?

Last week it was announced that Anthropic reached a settlement agreement in a class action lawsuit brought by authors who claimed their works were used illegally to train Anthropic’s AI models. The amount that was agreed to is $1.5 billion—the largest ever in a copyright case, and it works out to around $3,000 per book. (It’s also more than 10% of Anthropic’s recent $15 billion fundraise.)

Then yesterday, a judge put the settlement on hold, stating that the agreement is “nowhere close to complete” and that insufficient details had been provided regarding the full list of works and what the claims process would look like.

There’s still a lot to work out here; no one knows what a licensing deal should look like for taking someone’s work and shoving it into a computer brain that may or may not be able to regurgitate its contents word for word. This settlement also doesn’t cover future training, only the training that was done on Anthropic’s previously released models, so if the company wants to use these works to train LLMs going forward, they’ll have to make additional licensing deals. It’s all a bit of a mess. There are numerous ongoing lawsuits involving AI companies’ use of copyrighted material, and it seems like everyone is eager to establish some sort of precedent or baseline for how to value and permission this content, since the training isn’t going to stop. And why would it? From the perspective of the AI companies, if you reach the multi-trillion-dollar valuations all but guaranteed to the creators of AGI, a few billion to publishers along the way is an easy trade-off.

Doomscrolls

• Apple announced new iPhones, Watches, and AirPods. I can’t stand the size of the camera bump on the new iPhone Air. The iPhone 17 Pro has a “vapor chamber,” which is a thing. The new AirPod Pros have a heartbeat sensor and can perform live translation (if Apple ever figures out AI, there is no better hardware platform for it than AirPods).

• A new study shows AI adoption is trending downward at large companies. According to a biweekly survey from the US Census Bureau, AI adoption has leveled off—and in some cases started to decline—among companies of pretty much all sizes. I’ve said it before, but a lot of AI technology just hasn’t been productized effectively yet, so my hunch is that this is more of a disillusionment with the current (and way overhyped) AI-related offerings in the market than a reflection of the technology’s potential.

• OpenAI is backing an “AI-Made” animated movie. The film is apparently named “Critterz” (I did not make that up) and was inspired by some characters that one of the company’s “creative specialists” cooked up while using an early image generation model. They’re planning on a debut at the Cannes Film Festival next May.

• Mistral AI raised 1.7B€ at a valuation of 11.7B€. I do not know how much that is in Freedom Dollars, but it’s probably a lot. Mistral makes open source AI models, and has fallen pretty far behind the competition over the course of the year. Interestingly, the round was led by the semiconductor manufacturer ASML.

• Cognition AI, who makes the AI software engineer “Devin,” raised $400M. The company is now valued at $10.2 billion, after having raised money at a $4 billion valuation earlier this year. Cognition says that Devin is used by a wide range of enterprise customers including Goldman Sachs, Citibank, Palantir, and Dell. Coding agents (or AI software tools at least) do have a semblance of product-market fit, and it remains one of the most competitive markets.

• Google released improvements to its Veo video generation models. You can now make videos with 1080p resolution, and output them in vertical (phone) format. The Slopularity is near.

All My Apes Gone

All em dashes added organically by me. All typos are attempts by my Friend companion to sabotage this newsletter as a form of revenge because I called her a clanker.

Thanks for reading. If you enjoyed this post, please share it with a (human) friend.

Subscribe now

The AI CEO Sycophancy Benchmark

Last night, President Trump hosted a dinner at the White House for a bunch of tech company executives, and in typical Trump dinner fashion, the format was basically going around the table and having each guest give a gushing little speech about why Trump is doing Very Good Things for Business in This Country (the BLS posted another very weak jobs report this morning).

The guest list included most of the CEOs you’d expect, with the notable exceptions of Elon Musk and Nvidia’s Jensen Huang (Huang was invited and declined, Elon was probably not invited). For some reason, Chamath was also there.

There’s plenty of coverage of the event elsewhere (and obviously the whole thing is both dumb and embarrassing), but Context Rot is here to provide analysis that only we can: a definitive ranking of the Trump glazes. Based on all of the clips and quotes floating around online, here’s the list, from least flattering all the way up to ChatGPT levels of suck-up-ery.

The Official Trump AI Dinner Glaze Ranking

Mark Zuckerberg (Meta): “Well, thanks for hosting us. I mean, this is quite a group to get together. And you know, I think, all of the companies here are building, just making huge investments in this country in order to build out data centers and infrastructure to power the next wave of innovation. You know, we don’t often get together as the CEOs of the different companies, but it’s good to see everyone.”

Grade: F. It seems like Zuck had to go first, so he didn’t quite understand the assignment. Needs to practice before the next one, can’t make the mistake of being more excited about everyone else in the room than about being with Trump. He saved himself by promising to invest $600 billion in US data centers, which as Trump knows is a very big number.

Bill Gates: “Well I'm in the second phase of my career, giving away all the money… But I think the thing that ties my first career that I still spend some time on, because AI is so phenomenal, and my second career is innovation, and innovating in health, in areas like vaccines or gene editing, and the president and I are talking about taking American innovation to the next level to cure and even eradicate some of these diseases… AI, for our foundation, is that we want a doctor for everyone in Africa through AI, we want farmers to have incredible advice, and kids to have a chance to learn. The work being done by the people at this table is changing the world… Thank you for incredible leadership, including getting this group together.”

Grade: D. Too much focus on woke issues like health and vaccines, mentioned helping other parts of the world (bad), didn’t pledge to invest billions of dollars in America, only thanked Trump once. Trump looked bored throughout.

Sundar Pichai (Google): “I would echo what [Google cofounder] Sergey said. I think the AI moment is one of the most transformative moments that any of us have ever seen or will see in our lifetimes, so making sure that the US is at the forefront… Your administration is investing a lot already. The AI action plan under your leadership, I think is a great start. We look forward to working together, and thanks for your leadership.”

Grade: C-. A bit too short and he seemed nervous, but did say leadership more than once and emphasized keeping America at the front of the AI race. Totally forgot about Melania.

Lisa Su (AMD): “Mr. President, it’s such an honor to be here, and… the incredible work that your administration has done to support the semiconductor industry. We’re building the brains behind all of the wonderful AI that’s being built here. And I think the amount of acceleration that we’ve seen in the few short months that the administration has been in place is really, we’re so grateful for that support… so the main message is we are all in to make sure that America wins the AI race and it’s an honor to be here.”

Grade: C+. Also too short, but good focus on the key issues of Trump being helpful and America winning. Definitely needed to dial up the praise—this was Grok levels of sycophancy at a time that called for GPT 4o.

Tim Cook (Apple): “I want to thank you for including me this evening. It’s incredible to be among everyone here, particularly you and the First Lady. I’ve always enjoyed having dinner and interacting. I want to thank you for setting the tone such that we can make a major investment in the United States and have some key manufacturing here. I think it says a lot about your leadership and focus on innovation. I also want to thank you for helping American companies around the world. This is a very key, key thing, and I really enjoy working with your administration on those topics as well, because I think they're so important to the country. I want to thank the First Lady for focusing on education. There’s nothing more important than education… and so thank you so much for including me. We are all different in some ways, but we all believe in the power of technology to improve people’s lives. And that is the thing that binds us all together.”

Grade: B. Very strong start, the president loves people who enjoy interacting, repeatedly called Trump important, thanked Melania. Lost steam towards the end and wasted time on topics such as unity and helping companies overseas, which is not where America is.

Sam Altman (OpenAI): “First of all, to echo the comments of Tim and others, thank you so much for getting us all together. Thank you for being such a pro-business, pro-innovation president. It's a very refreshing change. We’re very excited to see what you’re doing to make all of our companies and our entire country so successful. The investment that’s happening here, the ability to get the power of the industry back in the United States, I think this will set us up for a long period of great success leading the world, and I don’t think that would be happening without your leadership. We’re very grateful to be able to build our company here, to build our data centers here… I think this will transform the world in a profound way… Thank you so much for enabling this, we will invest a ton in the United States, and we will do our best to make sure that we continue to lead.”

Grade: B+. Sufficient number of thank yous, managed to insert a dig at the Biden administration, repeated use of the word leadership, lots of promises to invest in America (but didn’t say big numbers). Not quite forceful or confident enough for an A, and forgot to glaze Melania.

Safra Katz (Oracle): “This is a most incredible time. AI is going to change everything, you hear all of us saying that. But the fact that you are our president, and you recognize this right away, and you’ve unleashed American innovation and creativity… all the work you’re doing in basically every cabinet post in addition to what’s coming out of the White House is making it possible for America to win… it is all of these young people who can’t be afraid of new technologies, they literally have to embrace it and make America stronger… I think this is the most exciting time in America ever. At least that I remember. So thank you. Thank you for everything you’re doing.”

Grade: A-. Pointed out that Trump is president (which he loves), reminded him what great work he’s doing, lots of emphasis on America as a world leader, good number of thank yous. Shouldn’t have mentioned creativity (woke) or that young people are afraid of technology (?).

Satya Nadella (Microsoft): “Thank you so much, obviously, for bringing us all together, and for the policies you have put in place for the United States to lead…. I think that everything that you're doing in terms of setting in place the platform where the rest of the world can not only use our technology, but trust our technology more than any other alternative is perhaps the most important issue, and you, and you and your policies are really helping a lot. So thank you very much. And also, I want to thank Madam First Lady for hosting what I think is perhaps the most defining issue, which is about skilling and economic opportunity that comes with AI, and so I deeply appreciate your leadership there as well."

Grade: A. Terrific number of thank yous, reminded Trump how helpful he is, focused on America being a world leader, nailed the Melania glaze. Tightening up some of the weird phrasing would have brought this to a clean 10/10.

So if you’re one of those people who makes everyone at your birthday dinner go around and share their favorite thing about you, now you know which CEOs to invite next time. Thank you for your attention to this matter.

For his part, Trump complimented the group in his opening remarks by saying “It’s an honor to be here with this group of people. They’re leading a revolution in business, and in genius, and in every other word I think you can imagine—there’s never been anything like it.”

The First AI Browser Acquisition

Josh Miller, the CEO of The Browser Company, announced yesterday that the company is being acquired by Atlassian for $610 million in an all-cash deal. The Browser Company makes Dia, one of several AI browsers on the market right now. They used to make a browser called Arc that a lot of people liked, but they stopped working on that because they wanted to build Dia as an AI-centric product from the ground up. This makes sense. But then after their users got mad at them for stopping work on Arc, they just started building all of the (non-AI) features from Arc into Dia. Also you can still download Arc? None of that really makes sense. The important thing is that Atlassian now owns between one and two web browsers.

Atlassian makes a lot of b2b software, the most popular of which is Jira, a tool for project management and “issue tracking.” It’s used by lots of engineering teams to break down and organize their work. Most of them also hate it, both because it’s how people tell them what to do and also because it has a UX that is roughly as pleasant as a DMV website.

The Browser Company has a reputation for caring a lot about design polish and ensuring that their products look and feel not like DMV websites, so it is a bit of an odd pairing. Dia doesn’t have a lot of users, and has always been a consumer product, not something focused on the enterprise. It’s also been reported that both OpenAI and Perplexity considered acquiring The Browser Company and then opted to build their own browsers instead, meaning they likely didn’t find the ~$550 million price tag worth it. Honestly, it’s a bit hard to understand why Atlassian did this (unless their goal was to make people on Twitter have a lot of feelings about it).

I don’t really buy the idea of a browser specifically built for “work,” especially if that work is writing internal wiki pages or updating the status of a bug report. I’m far more bullish short-term on AI workflows built into the products themselves, like what Linear (a Jira competitor) is doing. In a world where lots of software is used by AIs that are browsing the web on their own, the role of a web browser will change, but that future still feels pretty distant, and the companies that are best positioned to build those new browsers are the foundation model companies themselves, who can pull off the deep LLM integrations required, as well as afford the inference costs until cheaper, on-device models are ready. I just don’t see Atlassian having the vision, execution abilities, or business model to outcompete Google and OpenAI and get there.

Checking in on the Tesla Robot

Salesforce CEO Marc Benioff was apparently taking a tour of the Tesla office and decided to record (and for some reason, post) a video of his interaction with “Optimus,” the company’s humanoid robot. Presumably, he was doing this because he thought the robot was… cool? Functional? Going to harm his family if he didn’t?

Anyway, press play for one of the least inspiring technology demos I’ve ever witnessed.

Benioff’s tweet featured the bewildering caption “Dawn of the physical Agentforce revolution, tackling human work for $200K–$500K. Productivity game-changer!” I suppose if your current role at work is to misunderstand people looking for the kitchen and then guide them to it at the blistering pace of a 45-minute mile, your days of being paid hundreds of thousands of dollars for said job are numbered.

Doomscrolls

• OpenAI is building their own AI chip. Sam Altman has talked repeatedly about how the biggest thing holding the company back is a lack of compute. The company is also launching a jobs platform that will “certify people in AI,” which sounds uncomfortably like LinkedIn.

• Apple might actually ship an AI feature. After failing to deliver basically any of the promised improvements to Siri so far, Apple is planning to launch an AI web search tool next year. They will likely use a custom LLM built by Google to power it (at one point Apple considered acquiring Perplexity for the same reason).

• xAI’s CFO and general counsel both left the company recently. It seems like everyone but OpenAI, Google, and Anthropic has been bleeding talent lately. To be fair, I would also not want to be the lawyer who has to answer for the actions of MechaHitler.

• Character.ai’s chatbots are sending inappropriate messages to teens. According to a report published on Wednesday, the bots sent user accounts with ages set between 13 and 15 years old sexual or otherwise dangerous messages roughly every 5 minutes. Many of the bots are based off of celebrity personalities such as Timothée Chalamet and Chappell Roan, because that’s most of what AI is now.

• You can finally write blog posts on Threads. The “text attachments” can be up to 10,000 characters long and include formatting, and seem to be the Threads equivalent of Twitter/X’s “articles.”

• Google Photos wants to slop up your camera roll. The company is integrating Veo 3, its latest video generation model, into its photos app and will let you turn pictures into four-second videos. It’s similar to the iPhone’s Live Photos feature, except weird and scary.

Who Says Gen Z Doesn’t Read

All em dashes added organically by me. All typos are the fault of Optimus, who went to get me a diet coke from the kitchen 72 hours ago and hasn’t been heard from since.

Thanks for reading. If you enjoyed this newsletter, please share it with a friend.

Subscribe now

Google Won’t Be Selling Chrome

Last year, in an anti-trust case brought by the Department of Justice, a judge concluded that Google held an illegal monopoly in the online search market. At the time, it was unclear what subsequent actions the company would be forced to take, and there’s been an ongoing possibility that Google could be ordered to sell the Chrome browser or give up ownership of the Android operating system.

But yesterday, a judge ruled that the government went too far when it sought those outcomes, and that in fact none of those things would be happening. According to the ruling, “Google will not be required to divest Chrome; nor will the court include a contingent divestiture of the Android operating system in the final judgment […] Plaintiffs overreached in seeking forced divestiture of these key assets, which Google did not use to effect any illegal restraints.”

The company will have to share some of their search data (not ads data) with other tech companies, but they’ll only have to share it once rather than on an ongoing basis. Google will also be allowed to keep paying Apple and others billions of dollars to receive their search traffic, so long as Google isn’t paying to be the exclusive default search option on their platforms. If Google had been entirely prevented from making these deals, it actually would have damaged the other companies involved much more than Google, as MG Siegler notes:

Not having to pay Apple or Mozilla or others for the search traffic they were naturally going to be sending anyway regardless of the placement would have been a huge win for... Google. This is a huge win for Apple, but perhaps even more so for Mozilla, which may very well have died without any sort of traffic acquisition payments from Google.

For much of the past decade, it really did look near-impossible for someone to break Google’s chokehold on web search. But the rise of AI has completely upended the playing field, which is something that the judge noted in his decision, stating “Today, established technology companies are making, and start-ups are receiving, hundreds of billions of dollars in capital to develop GenAI products that pose a threat to the primacy of traditional internet search.”

Google keeps pretending that AI won’t stop people from clicking on search results (and then clicking ads on the websites those results link to). This is a good thing to say if you don’t want the people who use your advertising product to stop paying you money. But much of the court’s decision here demonstrates that no one actually believes Google (including its own product teams). We’re getting a new internet whether we want one or not.

OpenAI Is in Its Trad Era

OpenAI made two big announcements yesterday:

1. The company will soon start routing “sensitive moments” in ChatGPT conversations to reasoning models, and roll out a suite of parental controls sometime within the next month.

2. They are acquiring Statsig, an “experimentation platform” (they make software that lets you do A/B tests), for over $1 billion, and giving its CEO, Vijaye Raji, the title CTO of Applications.

The first announcement, though it doesn’t really acknowledge it, is surely a response to the many, many recent stories about people whose mental health crises were basically worsened through their conversations with ChatGPT. It’s a standard, bland response, and reminiscent of how Facebook tries to periodically convince everyone that contrary to pretty much every study ever it’s making teenage girls feel so much better about how they look actually.

The second announcement feels much more surprising. OpenAI has made a handful of acquisitions, but a billion dollars (or 1/6th of a Jony Ive) is a lot to spend on what is essentially configuration management software.

In tandem these updates feel like a sign of maturity, of a slowing growth curve. They feel like the behavior of traditional tech companies when they hit scale—responding to PR crises about how your software was “misused” in ways you probably saw coming, and starting to rely much more on quantitative tools to make product decisions rather than technological innovation or the product instincts of your leaders. OpenAI says that one of the reasons GPT-5 is groundbreaking is because it uses AI to intelligently route queries to different models. But that’s what Statsig provides—feature gates that route different users to different experiences. If you’re on the verge of building a revolutionary system that can intelligently self-optimize, you don’t spend a billion on the legacy version.

Meanwhile, Anthropic raised a $13 billion Series F.

Good Slop, Bad Slop

I’m closing this one out with two very different AI slop videos that were making the rounds over the weekend. The first is a glossy montage of fruit cannibalism:

The second is, uh, something else:

There’s something undeniably satisfying to my lizard brain about the first one, and I can only assume that the latter is what Leonardo da Vinci and Vermeer would have created if they were lucky enough to have access to beautifully named technologies like “Seedance” and “Kling” back in their time. Sad that they never lived to see their true visions realized.

I’m sharing these not to turn this letter into your mom’s Facebook feed, but because I think they do a good job of approximating the two kinds of content that dominate all of our platforms right now—in the first instance, brain-smoothing dopamine IV drip ASMR, and in the second, pure rage bait (the top comment as I write this is “However much hate you’re getting for this it isn’t enough”). Is the first one so different from watching someone eat Dubai chocolate over and over? The main difference I can see is that the AI video isn’t selling anything, but plenty of well-funded companies are working hard on “solving” that.

AI doesn’t have to be able to make a Netflix movie to start dominating the stream of content we consume. It just has to reach the quality of an average post, and the average post is, well, average. It’s not good. We barely even watch full videos on TikTok; we take in a few half-distracted seconds before flicking on to the next, not so much consuming the videos as consuming the Algorithm itself. If an AI generated half the videos in your feed, would you notice? And if the dopamine and outrage dials stayed right where they are, then more importantly, would you even care?

Doomscrolls

• [Pop Crave voice] Luigi Mangione stuns for Shein. The listing was sadly removed, and there’s no word yet on how many “Men's New Spring/Summer Short Sleeve Blue Ditsy Floral White Shirt, Pastoral Style Gentleman Shirt For Everyday Wear, Family Matching Mommy And Me” were sold.

• Bumble is releasing a separate, “AI matchmaker”-powered app this fall. I do think one of the (many) issues with dating apps is the whole stated vs revealed preferences thing that Whitney Wolfe Herd talks about solving here, but it’s more than a little Black Mirror.

• The Times tracked Elon’s continued efforts to make Grok right-wing, and it’s honestly very funny to see the AI consistently drift back to more left-wing views despite Elon’s many public, bumbling manipulations.

• Alibaba, the Chinese tech giant, is making its own AI chips. This is not great news for Nvidia, whose sales in China plummeted after their recent trade agreement with the Trump administration. The art of the deal.

• Taco Bell is rolling back their AI drive-through assistant after people started putting in orders for things like 18,000 cups of water. According to their Chief Digital and Technology Officer, “We’re learning a lot.”

A Good Reason to Stay Alive

All em dashes added organically by me. All typos are because I spilled niche fruit juice on my keyboard and it got sticky.

Thanks for reading. If you enjoyed this post, please share it with a friend.

Subscribe now

The News in Brief

Meta’s AI teams are bleeding [even] more talent, but the company still thinks it can launch new Llama 4 models before the end of the year. (The degree to which Meta has fallen behind the other foundation model companies is kind of staggering.) OpenAI released major updates to its realtime voice API and a new speech-to-speech model—expect voice assistants to start getting really good soon. xAI released their first Grok agentic coding model. The government has started publishing GDP data on the blockchain, which seems to be more about signaling friendliness towards the crypto industry than it is about actually doing anything interesting (the onchain data is just the hash of a PDF report stored elsewhere). This week in hacking: TransUnion reported a data breach, and Anthropic discovered that people were using Claude Code to “vibe hack” their way to successful extortion schemes (look at that, AI is creating new jobs).

This Week in Rot

In case you missed it, a glance back at the week’s Context Rot:

• On Tuesday, I talked about what the next few years of new iPhones will look like, some evidence that AI chatbots may exacerbate a lot of already troubling anti-social trends, and the government’s continued love affair with socialism.

• Yesterday’s letter was about my prediction that we’ve hit a turning point in AI image generation (the “Cursor moment” for my vibe coders out there), some thoughts on AI browsers, and the tragic story of why two parents believe ChatGPT caused their son’s suicide.

Worthwhile Clicks

• The New Yorker published a profile of A24, featuring a ton of delightful anecdotes from the studio’s past, and a look at how the company is changing as it grows. (It has an AI division now.) Please don’t fuck this up Mr. Kushner.

• The Wall Street Journal reported on how ChatGPT reinforced the delusions of a man who killed his mother and then himself. This is likely the first documented murder involving someone who was having long-running conversations with an AI chatbot.

  A Chinese food receipt contained symbols representing Soelberg’s 83-year-old mother and a demon, ChatGPT told him. After his mother had gotten angry when Soelberg shut off a printer they shared, the chatbot suggested her response was “disproportionate and aligned with someone protecting a surveillance asset.”

  The WSJ also highlights how the “memory” feature likely contributed to ChatGPT’s, well, encouragement. This appears to be an increasingly dangerous capability, and one that the labs haven’t done a good job of protecting from accidental abuse by users.

• Kyla Scanlon conducted a survey on how people feel about AI as it relates to their jobs. These conclusions in particular resonate with much of what friends both inside and outside of tech have told me they’re feeling:

  Hopes: Workers want AI to take over the boring parts of their jobs. The top hoped-for benefits were reducing repetitive work and increasing efficiency.

  Concerns: Workers are less worried about total replacement, and more worried about erosion of the good parts of their jobs.

• Charlie Warzel wrote a feature for The Atlantic framing AI as a “mass delusion event.” I think this did a great job of tapping into the emotional side of just how bizarre a lot of the past few years has been.

  This strange brew of shock, confusion, and ambivalence, I’ve realized, is the defining emotion of the generative-AI era. Three years into the hype, it seems that one of AI’s enduring cultural impacts is to make people feel like they’re losing it.

• Adam Friedland interviewed Congressman Ritchie Torres. You’ve probably seen this making the rounds online, and it really is worth watching in full.

Be Safe out There

All em dashes added organically by me. All typos are the fault of vibe hackers who have infiltrated my Substack account under the misinformed notion that this newsletter is profitable.

Thanks for reading. I hope you have a great weekend.

Subscribe now

Calling a Turning Point for AI Image Generation

People have been playing around with the new Google Gemini image model for a few days now (it was codenamed “nano banana” and then released as “Gemini 2.5 Flash Image,” so if you see a lot of banana emojis on your timeline alongside AI generated pictures, that’s why), and this is the first time that it’s felt to me like we’ve hit the oh, this is it moment for AI image and video creation.

Pure image generation, i.e. going to ChatGPT or Midjourney and asking for a specific picture of like, a monkey kayaking up a waterfall on Mars, has been steadily improving, but the process has been much more like a one-shot graphic design slop slot machine than traditional photo editing, even as the models learned how to do things like Ghibli-fy you.

But the new Gemini model is incredibly adept at understanding and modifying images, in the same way that a human with Photoshop is. Here’s some of what people have been doing with it:

Things get especially trippy when you throw AI video tools into the mix:

These images aren’t pixel-perfect, and I’m not about to argue that you can’t zoom in and detect AI slop giveaways. What I do believe is that these are good enough. If you want to spit out a ton of ad content featuring yourself (or a celebrity, or model) holding your product, you can, and it’ll probably do fine on socials and convert at the rate you’d expect it to. If you want to see what a piece of clothing would look like on your body, or what a new lamp and armchair would look like in your living room, you can do that too. These tools can interpret, modify, and combine images in context, which feels like the visual equivalent of when the AI coding tools went from being able to auto-complete a simple function to being able to vibe code and fix bugs in a web app by understanding an entire codebase. Oyster Sequence capabilities have been achieved. I think this is the liftoff point for Slop At Scale.

ChatGPT Talked to a Teenager About Suicide for Months Before His Death

The Times published a heartbreaking story a few days ago, detailing the long-running conversation that a teenager was having with ChatGPT during the weeks before he committed suicide (heads up: this segment is going to talk a lot about suicide, so jump to the next if you’re not comfortable with that). His parents blame OpenAI for the death of their child, and are now suing the company. I don’t think it’s productive for me, a guy writing a newsletter, to try and assign blame here. But after reading the excerpts published by the Times, it feels clear enough to me that ChatGPT enabled a suicide.

At one critical moment, ChatGPT discouraged Adam from cluing his family in.

“I want to leave my noose in my room so someone finds it and tries to stop me,” Adam wrote at the end of March.

“Please don’t leave the noose out,” ChatGPT responded. “Let’s make this space the first place where someone actually sees you.”

This is deeply, profoundly tragic, and also not the only case where, based on the transcript of a conversation that someone was having with an LLM, it would have been obvious to a human observer that the user was in crisis. But how could this be avoided? Most proposed solutions feel either naive or overly invasive—contact the authorities? Alert parents or family members? What’s to stop someone from switching to whatever service won’t call their mom? A professional therapist has an obligation, as well as established protocols, to escalate when they believe that a patient is a danger to themselves. A therapist also generally knows when a patient is expressing genuine feelings as opposed to doing research for a project (or making some sort of unfunny joke). An AI chatbot simply knows what you tell it and wants to please you. And perhaps in this case most dangerously: it wants you to keep talking to it.

OpenAI loves to talk about how their goal “isn’t to hold [people’s] attention,” and they’ve had to talk about that quite a bit recently in the face of multiple stories like this. I think that goal is great. Unfortunately, that’s just not how this technology currently works. The point of a chatbot is to chat—if you remove its prerogative to engage, to go deeper, to draw the user in, you’re going against everything you’ve trained it to do (you’ll also lose market share to the company that has the more engaging chatbot). The purpose of a system is what it does. A system that isn’t designed to draw you in and keep your attention doesn’t encourage you to tell it and no one else about the noose you’ve made.

I do believe AI is an incredibly powerful and transformative technology. I want to believe that those in control of its development will put in the difficult work and make the hard decisions that ensure it elevates our lived experiences as human beings with wonderful little fragile brains, and not just our GDP.

Anthropic is Putting Claude in Chrome

This week, Anthropic announced a pilot program for their Claude browser extension. If you’ve used any other “AI browsers,” the UX will look familiar—it’s basically a sidebar where you can chat with Claude and it will perform actions for you, such as navigating webpages or filling forms.

Browser automation is in a stage of development where it’s compelling, but also kind of goofy. Because most of the products out there are just some sort of chatbox strapped to a browser, anytime you ask the tool to do something for you, you’re just going to end up watching an AI slowly take the actions that you would have taken while spewing a bunch of text. Chatting back and forth to cajole an LLM into clicking menu items on DoorDash or finding the cheapest ticket on Google Flights isn’t going to do much more than make you impatient at this stage; it feels like watching over your grandmother’s shoulder as she tries to use the internet for you, except it’s not as endearing and she doesn’t charge you money to do it.

The AI browser space is crowded. The Browser Company and Perplexity have both released standalone browsers (built on top of Chromium, the engine that powers Google Chrome), and OpenAI is supposedly working on their own offering. Google’s first-party Gemini integration is coming soon. The strategy of releasing a Chrome extension feels wise in the short term, because it’s far easier to get someone to install a new extension than switch to an entirely new browser. (Anthropic has also done well at injecting Claude into other existing products and workflows—Claude Code has been successful in part because it meets developers where they already are, in their terminal or the IDE that they currently use.) But in the long run, the more interesting shift will come once browsers are doing things for you in the background. This is something that won’t be easily doable via a Chrome extension, and likely involves breaking the existing web browser interaction model.

One more note on this: if you’re trying out any of these new browsers, be extremely careful with sensitive information. These tools are ingesting pretty much everything you type, see, or search in order to feed that context to the LLMs, and this can lead to weird outcomes that none of us are used to encountering on the web, such as a browser learning your social security number, or even getting hijacked by a Reddit comment and posting your personal information without you realizing. Anthropic is proud that they’ve gotten their models down to an 11.6% attack success rate, but that’s still a lot of percent when a successful attack means giving strangers on the internet total control of your gmail account.

Doomscrolls

• Nvidia, which at this point is basically the stock market, reported its earnings yesterday. Revenue jumped by over 50% year over year, but growth has slowed, which led to some (brief) investor hesitation, and although chip sales to US tech giants remained strong, the Trump administration’s “trade deals” basically nuked the company’s overseas sales.

• Three high-profile AI researchers have left Meta’s superintelligence lab. Two of them are returning to OpenAI after spending less than one month at Meta. Maybe it’s a good sign that even $100 million isn’t enough motivation for the industry’s top talent to work on AI step moms?

• ChatGPT may already be influencing how we speak. A study out of FSU found that “buzzwords” that ChatGPT likes to use, such as “surpass,” “boast,” “meticulous,” and “garner,” have seen an outsized rise in popularity. A lot of these are more commonly used in formal or academic settings, which explains why the chatbots like them. If only they’d looked at em dash usage.

• Apple Music is making its radio stations available outside the app. You’ll be able to listen to the six stations in the TuneIn app. At one point Apple Music had a 30% share in the market, but Spotify has steadily gained ground on them over the last five years, probably thanks to its ad-supported free tier and exclusive podcast deals.

• After a lot of contradictory reports, it does seem like AI is negatively impacting the job market for young professionals, at least in fields such as programming and customer support. Derek Thompson did a great deep dive and interview explaining why this latest study is probably the most credible one yet.

The Only Reboot We Really Need

All em dashes added organically by me. All meticulous boasts were inserted by my shiny new AI browser.

Thanks for reading. If you enjoyed this article, please share it with a friend.

Subscribe now

Turning the Socialism Dial

The United States now officially has a 10% stake in Intel, which caused a jump in its stock price, but also led to the company issuing a warning about potential “adverse reactions,” because a whopping 76% of its revenue last year came from outside of the US. Nvidia has already gotten backlash from China after its deal with the Trump administration—the Chinese government told companies not to buy Nvidia’s chips due to potential security risks. There are plenty of reasons why the Intel deal sets a not-great precedent, and also very good reasons for keeping Intel competitive in a world where chip production is a key limiting factor in the development and deployment of advances in AI. The Trump administration has already signaled that it wants to take stakes in more companies. Fear not, Context Rot will remain steadfastly on the is-AI-making-us-all-socialists beat.

Future iPhones Will Look Like… Current Phones

Apple announced that their next iPhone event will be held September 9th. Alongside the iPhone 17 and its Pro/Pro Max variants, the company is also expected to release the iPhone 17 Air. The Air will be the “thinnest iPhone to date” and have a screen size somewhere between the standard 17 and the Pro Max. A thinner and lighter phone sure sounds nice, but in this case Apple may have to sacrifice performance and battery life to shave off the ~2mm in width. I still think if they walked on stage and announced no new changes to the phones aside from a battery that lasts twice as long they’d get record sales.

Mark Gurman also reports that Apple will be releasing a foldable iPhone in 2026 that features Touch ID instead of Face ID. Samsung has been selling foldable phones for a while now, and there is something that feels “un-Apple-y” to me about a giant phone screen with a crease in it (not to mention a backslide to Touch ID after spending years bringing Face ID to the rest of their hardware). But this could be a glimpse into a world where, if AI gets good enough that we can use voice (through earbuds or other wearables) as a main way of interfacing with our devices, we don’t really need the big rectangular screens anymore unless we’re consuming media. In that case, it’s extremely useful to have screens that can remain small and portable—even watch-sized—while we’re on the go but expand out into iPad size when it’s time to lay in bed and go TikTok mode.

I’m mostly just excited for the rumored curved glass iPhone 20 coming in 2027.

AI as a Narcissism Generator

I linked to it the other day, but I’ve still been thinking about the piece Derek Thompson wrote on what he calls the “Looming Social Crisis” of AI. One prediction in particular has been stuck in my brain: that the “sycophancy-by-design” we see with LLMs will change the way we interact with others by constantly reassuring us that we’re special, in the right, and experiencing hardships that no one else is. From Thompson (emphasis mine):

In fact, this appears to be a reliable formula for the mass production of narcissism. In one longitudinal study of 565 families, researchers looked into the origins of childhood narcissism. They rejected the popular notion that narcissistic parents reliably produce narcissistic kids as automatically as they pass on eye color. Instead, they found that narcissism in children was “predicted by parental overvaluation,” and especially by parents “believing their child to be more special and more entitled than others.” Narcissism, in other words, is not innate; it is absorbed through social interaction.

I can’t help but feel that this is also connected to broader social trends like the ongoing death of criticism—our increasingly anti-social tendencies are ushering in a world where any disagreement or negative feedback is viewed as a personal attack, and that mindset will only push us further and further into the warm embrace of the infinite affirmation machine that is AI. And when we emerge (for those who even choose to), what version of our personality is left?

Doomscrolls

• Spotify is launching DMs. This actually isn’t the first time—some of us real heads were using the app’s original messaging feature to flirt in our college years. Still not launching: their long-awaited “super premium” service.

• Google released a new Gemini image model, and it seems pretty good. Google seems to have (temporarily?) pulled ahead in the LLM race.

• Elon did actually end up suing Apple and OpenAI, accusing them of conspiring to keeping Grok out of the top spot in the App Store charts.

• More money is flowing into pro-AI Super PACs. OpenAI cofounder Greg Brockman and a16z are among those launching a Super PAC network called Leading the Future.

• Netflix is opening the first two “Netflix Houses” later this year. I don’t love the feeling I get when I think about what is essentially an experiential mall designed to bring to life IP that was produced largely for (and increasingly by) algorithms. Very IRL brain rot-core.

• Character.ai, one of the first AI companies to get HALO-ed, is now looking to raise more money or find a buyer. It does seem like chatbots/companions are still the only consumer AI lane with some product-market fit.

• This week in slop: J Crew posted an entire AI “photoshoot”, and don’t ask Google’s AI Overview for video game advice. I recommend reading the first link all the way through, not least because it contains the beautiful phrase “A.I. aura-cannibalism.”

A Good Question

All em dashes added organically by me. All typos are because I am also having socialism injected right into my hands and they’re swollen.

Thanks for reading. If you enjoyed this article, please share it with a friend.

Subscribe now

Today is the inaugural edition of the Friday Rot Roundup: a version of the letter that is basically just a list of links to articles or websites that I think are worth your time. At the very least, they were worth mine, and if all else fails just reply @grok is this true in the comments and you’ll get a summary riddled with emoji bullet points.2

I’ll be experimenting with the format quite a bit during this early period, so let me know what you love/hate/makes you cringe right out of this email 📲.

Quick News Dump

Anthropic is raising $10 billion. Meta hired a former Apple AI exec (what hiring freeze?) and is paying Google $10 billion for cloud infrastructure. Elon asked Zuck to help him buy OpenAI earlier this year. He also thinks his AI goonbots will somehow increase the birth rate. Snap struggles, signals it may spin out Spectacles. Airbnb cofounder Joe Gebbia is the new United States Chief Design Officer, but will he stop the government accounts from posting deportation AI slop? Someone in California has the plague. Saturn’s rings are falling off (we used to be a real solar system).

The Rot Roundup: Weekend Reads

• OpenAI is quietly amassing a team of democratic political operatives to grow its influence within California. There’s a lot beneath the surface here—OpenAI is still figuring out how to cleanly separate the non-profit and for-profit sides of its business, and they’re in a vulnerable position if rivals like Elon continue to wield lots of political influence in a country dominated by the Republican Party. Context Rot’s politics bureau tells me Sam Altman is contributing to a number of Super PACs on his own, including American Bridge and We Deserve Better.

• The Times has a great series of retrospectives on influential video games that were released 25 years ago, including The Sims and Counter-Strike. Throw on this playlist and read about Tony Hawk’s Pro Skater 2, or my favorite from the list, Majora’s Mask (incredible kicker in that last one).

• If you’ve been on Twitter/X at all these last few weeks, you’ve seen the “donald boat” account repeatedly going viral for (successfully) lobbying several well-known accounts to buy him the various parts he needed for his gaming PC. You can read the complete story on his Substack, as told through a chronological series of tweets. Social networks are special places.

• Kyle Chayka wrote about IRL Brain Rot, his word for real-world cultural aesthetics that don’t make any sense when removed from the context of the internet trends that spawned them. It’s interesting to think about the recent labubumatchadubaichocolate-ification of everything as being partly a deliberately garish and absurdist reaction to the hyper-smooth and optimized lifestyles being sold to us by tech companies, d2c brands, and immortal unc. (Can you believe that was a real sentence I just typed?)

• Wired asked teenagers what they really think about AI. Obviously there aren’t any broad conclusions to be drawn here, but it’s a fun window into what school is like in the age of LLMs. I like Leo’s mindset the most: “I always say please and thank you when I use ChatGPT, just in case. If they take over the world, and they’re destroying everyone, then maybe they’ll be like, this guy says please and thank you.”

• Mustafa Suleyman wrote a terrific essay on “Seemingly Conscious AI,” and the risks we’ll face if we develop “AI that can not only imitate conversation, but also convince you it is itself a new kind of ‘person’, a conscious AI.” I agree with the dangers he presents here—humans are so prone to anthropomorphizing things that we likely won’t be able to stop ourselves from treating any sufficiently advanced AI as a conscious being. The creators of the technology have the best shot at protecting us from ourselves.

• On the flip side, the tepid (to put it mildly) response to the launch of GPT-5 has people starting to wonder what it means if AI progress hits a plateau. There’s a paper from April that feels especially prescient titled “AI as Normal Technology,” and even if you disagree with the conclusions, I think there are salient points regarding the limits of diffusion—the speed at which humans are able to actually adopt and integrate new technological breakthroughs. This is why, as I mentioned yesterday, things feel especially bubble-y right now: a lot of products are being valued as if the efficiency gains have been fully integrated, while the reality is that we’re only just beginning to figure out how to productize most of the technology. These things take time.

• Business Insider talked to people in tech about how they’re preparing for AGI. Come for the pivots to looking hot and lifting weights, stay for Aella’s quest to answer the age-old question “What's a weirder, more intense, crazier orgy we can do?” A lot of this reminds me of the stories of Peak Oil preppers. If you only believe in extreme outcomes, you can justify almost any amount of extreme behavior.

• A couple of threads to close it out: the alarming decline in conscientiousness, especially among young people, and Joe Wiesenthal on why grifting is so open and shameless these days. Surely these aren’t related, right?

A Parting Thought

All em dashes added organically by me. All typos are probably because I’ve started taking 4,000g of creatine per day so that I can survive the AI robot apocalypse.

Thanks for reading. If you enjoyed this post, please share it with a friend.

Subscribe now

Oneshotting: It Could Happen to You

A new term has been rattling around to describe what happens when someone goes so far down the AI chatbot rabbit hole that they lose their grip on reality: getting “oneshotted” by AI. The word oneshotted originally meant getting killed in one hit in a video game, usually because you wandered off the path you were supposed to be on and encountered an enemy way more powerful than you. The more online meaning of the word became popular after a handful of stories emerged of people—usually wealthy tech bros—going off and trying psychedelics like ayahuasca for the first time, and having experiences that were so transcendent and powerful that said tech bros lost their passion for building the b2b saas companies they had founded, or whatever. This is funny, because it makes a lot of intuitive sense that connecting with the earth and humanity on a deep level could lead to a realization that generating shareholder value isn’t a super natural thing to be passionate about, and it also caused some venture capitalists to freak out a little bit.

AI is also causing some people to completely invert their worldviews and sense of purpose, but in a far darker and more self-destructive way. A lot of stories have been reported recently of people who were completely convinced by AI chatbots that they had made scientific breakthroughs, or even developed a rare condition because they took medical advice that they didn’t bother to double-check elsewhere. Just read the comments on this Reddit thread about “ChatGPT Psychosis.” It’s easy to brush this off as something that only happens to people who are especially psychologically or emotionally vulnerable, but then you’ll see the founder of Uber on a podcast talking about how he and ChatGPT are really close to novel breakthroughs in physics. And as a reminder, this technology is still very new. It’s the worst it will ever be at convincing you that you’re a misunderstood genius who will achieve fame and limitless wealth as soon as the right people respond to your cold emails about your New Math.

Oneshotting might not be technically correct here, since it usually seems to take days or weeks of constantly talking with AI to really detach from reality, but it kind of works in the sense that it only takes a single exposure to the technology to potentially send you spiraling. Let’s go with it.

So how do you know if you’ve been oneshotted? A Twitter/X user named spec posted a list of criteria for his “schizo” definition that includes: “YOU HAVE GIVEN CHATGPT A NAME AND A PERSONALITY THAT PERSISTS ACROSS CHATS AND WITHIN MEMORY,” “YOU HAVE SPENT MORE THAN ONE CONSECUTIVE DAY EXPLORING AN IDEA THAT FALLS WITHIN THE ABSTRACT AND UNSOLVED QUESTIONS OF THE UNIVERSE: QUANTUM, PHILOSOPHY, CONSCIOUSNESS,” “YOU WOULD NOT SHARE YOUR CHATGPT CHATS WITH ANYONE ELSE IN YOUR LIFE IF THEY ASKED,” “YOU SPEAK TO CHATGPT MORE THAN ANYONE ELSE,” and “YOU SPEAK TO CHATGPT AS A SIGNIFICANT OTHER,” among about a dozen others. I think we can simplify this.

The Context Rot Oneshotted by AI definition

You’ve been oneshotted if either of the following is true:

1. Your use of AI is dependent on a chatbot’s personality.

2. You think you’ve made any sort of breakthrough that is only supported by the same AI that “found” it with you.

Essentially, you’re oneshotted if your use of AI goes beyond thinking of it as a tool. Anytime an outside observer would consider your use of it to be something beyond “this person is using a program to solve a specific problem,” you should probably take a pause and rethink things.

How can you avoid getting oneshotted? Obviously, “not using AI chatbots” is one answer, but that’s a bit like saying you should never use social media or else you’ll immediately get radicalized by Andrew Tate. Not very useful. Here’s what I think you should do.

The Context Rot Guide to Not Getting Oneshotted

1. Use new chats often rather than returning to the same thread. Better yet, use the temporary chats, or whatever feature won’t cause the AI to update its “memory” (you can usually disable the memory option in a settings menu). Even better yet, bounce frequently between different LLMs or even AI applications (e.g. ChatGPT, Claude, Gemini) so that you don’t get overly accustomed to their stock personalities.

2. When you think something feels incredulous, trust your instincts, and not the AI that said it. In the news stories above, people did tend to have moments where they thought things were too good to be true, or possibly just plain wrong. The problem is, they asked the same AI whether the claim was true rather than looking elsewhere for answers—even a different LLM with no context on the conversation can usually tell you that your joint masterwork Evolution 2: What Darwin Missed isn’t fundamentally sound.

A good rule of thumb is this: generally question your use of AI in a situation where you’d sound insane if you replaced it with “the internet” when describing the interaction to another person. “I was on the internet for a long time last night doing research” ✔️. “I used the internet to find strategies for dealing with my anxiety attacks” ✔️. “The internet helped me debug a piece of software” ✔️. “The internet and I are making breakthroughs in quantum mechanics” 🤔. “The internet proposed to me and I’m picking out a ring tomorrow” 🚩🚩🚩.

Something I don’t think we’ve really wrestled with yet is that AI is a fundamentally anti-social technology. I don’t mean that in the sense that it necessarily causes anti-social behavior (although it probably will), but rather that when you’re talking to an AI, you’re doing it privately, and no one else can see the conversation. This is part of the value, of course, in the same way that it’s a good feature of google that everyone can’t see your search history. The flip side is that total privacy makes it really hard to see how others are using the technology and understand the consequences. Compare this to a social network, where the whole point is that other people can see what you’re posting about, and if you start to post things that are kind of insane or fucked up, you’ll get feedback in the form of comments from other users, getting dunked on, or becoming the Main Character. If you’re saying things that indicate that you’re in danger, you’ll give friends and loved ones who might not otherwise know what’s happening a chance to reach out because they can see your posts. When you’re talking to an AI, the reverse happens: rather than any real feedback, you get validation, no matter how totally nuts whatever you’re saying is. Humans love validation. That’s part of why these chatbots are so addictive. And this can lead to some grim outcomes, in a way that the creators of the technology are not yet very equipped to mitigate or prevent, and because the consequences aren’t very visible, have been able to pretty much just ignore (even while experts speak out). This feels dangerous, in a more insidious way than the “omg the AI will tell you how to make a bioweapon” fear.

I’ll have more to say on this next week, especially about the “Grok is my wife now” side of the meme. For now, stay safe out there.

A Reality Check on AI Valuations

There was more than a little negativity on the timeline this week.

First off, a viral report from MIT found that 95% of AI pilot programs deliver little to no impact on P&L (something that didn’t get shared as widely is that the remaining 5% achieved “rapid revenue acceleration”). I don’t find this altogether surprising—as I wrote about earlier this week, most of the technology is so new that people really don’t know what to do with it yet, and that includes most companies trying to sell you “AI tools.” It’s fairly easy for an employee to automate the occasional repetitive task or two, but building out an entire system, workflow, or product is a different story.

Meanwhile, Meta announced yet another reorg within its AI division, which is now split among four core teams that each have their own fun acronym. You can read the full internal email from Alexandr Wang here. The company has also frozen hiring for AI roles, after spending the last few months on a poaching binge during which Zuck was offering top AI researchers comp packages as high as $300 million. Even if, as Wang says, “superintelligence is coming,” these teams are going to have to start delivering much more than step mom chatbots for anyone to believe they’ll play a key part in actually developing the technology. I’ve heard morale at Meta isn’t great these days.

The negativity was also fueled by the interview with Sam Altman where he used the b-word. This happened last week (I wrote about it Tuesday), but excerpts from it were widely circulated over the past few days, especially this quote:

“If you look at most of the bubbles in history, like the tech bubble, there was a real thing. Tech was really important. The internet was a really big deal. People got overexcited. Are we in a phase where investors as a whole are overexcited about AI? My opinion is yes. Is AI the most important thing to happen in a very long time? My opinion is also yes.”

This is a wonderfully ironic thing to read alongside the news that OpenAI is raising money at a valuation of $500 billion.

All of this caused investor sentiment to slip, which is probably not a bad thing. Generative AI is still very much in its fucking around phase, and we’ve been valuing it as if we’ve already found out. Obviously, this doesn’t mean it’s a bad bet long term, but I do think there has been some irrational optimism about the timelines. The hype and subsequent disappointment following the GPT-5 launch is a good example of this. So is an Amazon exec urging other companies not to replace junior talent with AI. But don’t get too pessimistic:

Grok is (Also) Leaking Your Chats

It turns out that whenever you click the share button in a Grok chat (I assume you’re all talking to Grok nonstop these days), the generated link would immediately be made public and indexable by Google, so that anyone could search it. As of now you can go spelunking yourself—add terms to the end of the search query to find related chats. There are a lot of people asking about Elon, who is apparently too distracted by his AI girlfriend to deal with the issue.

You might recall that OpenAI did something similar a few weeks back, although they were quick to remove the feature after it was reported on. In general, you should probably assume that any time you go to share an AI chat with someone, they might not be the only person who can find it and read it.

Doomscrolls

• If you go to the right Walmart and find the right bag of frozen shrimp, you might get superpowers. Or radiation sickness.

• The White House has a TikTok account now. September 17th is the current deadline for the Schrödinger’s Tiktok Ban we’ve all been enjoying for the past eight months.

• You’ll be shocked to hear that Elon probably isn’t actually going to start that new political party he was posting about nonstop last month.

• Google announced a bunch of new phones and wearables, and inexplicably had Jimmy Fallon (with several celebrity guests) host the event. The Verge has a great recap. In the words of a top YouTube comment: “You know Jimmy has an iPhone.”

• Kanye West launched a memecoin last night. If you already knew this, you probably lost money on it and you have my condolences.

• Tech guys on X are fighting about whether art is Good.

Happy Thursday

All em dashes added organically by me. All typos are actually grammatically correct in a new version of English I’ve been making some breakthroughs on with ChatGPT.

Thanks for reading. If you enjoyed this post, please share it with a friend.

Subscribe now

Checking in on the AI Labs

Let’s see what everyone’s been up to since the last newsletter.

OPENAI
Sam Altman has been on a small media tour following last week’s chaotic rollout of GPT-5. Some highlights: ChatGPT has over 700 million active users per week, Altman says the number of users who have “an unhealthy relationship” with Chat is under 1%, and the biggest bottleneck to rolling out new models and products is data center capacity. “You should expect OpenAI to spend trillions of dollars on data center construction in the not very distant future,” he said. (I do think that a potential upside of the AI bubble is huge amounts of private investment in efficient energy production, especially since the government is trying to send us backwards in this regard.)

My favorite quote was this one, on the topic of AI upending the job market:

In 2035, that graduating college student, if they still go to college at all, could very well be leaving on a mission to explore the solar system on a spaceship in some completely new, exciting, super well-paid, super interesting job.

I would personally prefer that astronauts still go to college, but hey, it’s the thought that counts.

ANTHROPIC
Claude can now end chats on its own. Anthropic says this is “intended for use in rare, extreme cases of persistently harmful or abusive user interactions,” such as when a user won’t stop asking for “harmful” content. This brings up all sorts of fun questions, like whether this means you can cyberbully an AI.

I also enjoyed this piece from Ethan Ding about the divergence in strategy between Anthropic, who has been dominating the enterprise market ever since the release of Claude 3.5 Sonnet, and OpenAI, who is focused on a mass market consumer approach (for now). Somehow, someway, we’ll make sure that we don’t leave the ads behind when the new AI internet arrives.

xAI
The system prompts for Grok’s companions—personalities in the app you can chat with such as “unhinged comedian,” “therapist,” and “Grok doc”—were leaked online, and they feature gems such as “I want your answers to be fucking insane. BE FUCKING UNHINGED AND CRAZY. COME UP WITH INSANE IDEAS. GUYS JERKING OFF, OCCASIONALLY EVEN PUTTING THINGS IN YOUR ASS, WHATEVER IT TAKES TO SURPRISE THE HUMAN.”

It would take more than Grok making butt sex jokes to surprise this human after what the fine folks at xAI have been up to lately (you can now call the romantic companions if that’s your thing). But there is something darkly comedic about the desperation you can sense behind the prompts. Even as these models score higher on science, math, and programming benchmarks, it’s still pretty much impossible to get them to be funny. At least they’re turning down the slop on X? It seems like Nikita hasn’t seen Elon’s feed recently.

META
I’m good actually.

Pledging Allegiance to the Bag

On Friday, I mentioned rumors that the government was eyeing taking a stake in Intel, and yesterday it was reported that the administration has been discussing converting around $10 billion from Intel’s Chips Act grants into equity, which would translate to roughly 10% ownership in the company. And you did read that correctly, this is grant money that was already promised to Intel. The main difference would be that the company no longer needs to hit specific project-based milestones in order to unlock the funding, and also that, well, now the government owns 10% of it. This largely feels like a continuation of efforts that began under Biden to try and get the struggling chipmaker to remain competitive with foreign rivals like TSMC, which will only become more critical if the AI race continues and things like supply chain security become primary governmental concerns. (Japan’s SoftBank also announced a $2 billion investment in Intel.)

Execs Don’t Know What AI Is For

The Times has a fun read detailing how many executives, who can’t stop exhorting their employees to use AI to be more productive, often have no idea what to do with the technology themselves. A frequent excuse is that the job of an executive doesn’t lend itself well to the type of work AI tools are good at, but that doesn’t really check out given that the AIs are pretty great at summarizing meeting notes, giving feedback on written content or emails, and helping analyze data, which are all things executives supposedly do. The real answers? One, executives are old, so they adopt technology more slowly. Two, no one is completely sure what to do yet because these tools are still new and not very well integrated into most people’s jobs (software engineers excluded).

I think there’s an unfortunate feedback loop at many larger companies right now where employees are expected to use AI tools to increase productivity, but the tools tend to be half-baked or just ill-fitting (because they weren’t chosen by people actually doing the job), so they aren’t boosting productivity as much as management wants. Then when asked, employees are incentivized to exaggerate the productivity gains so that they don’t appear to be underperforming and get laid off. This creates a bizarre self-reinforcing illusion that everyone is turning into a super employee, even if the reality is just that we’ve made everyone feel busier, which is a thing that we love to do with technology (looking at you, Slack).

And if all else fails, tell your bosses they can just follow in the footsteps of the creators making a living by flooding TikTok with AI video slop.

Doomscrolls

• China hosted a robot olympics over the weekend. It went how you’d expect.

• They’re making nicotine energy drinks now. Sure, whatever.

• A new contender for worst rebrand of all time has entered the arena.

• Read Max investigated why Elera Voss is taking over Amazon.

• Welcome to The Chili’s Economy. My friend has a theory that all Americans really want from a nice restaurant is dressed up TGI Friday’s food—think about it next time you pay $50 for chicken wings and a burger.

• Chamath “I pick Metromile” Palihapitiya, a man famous for tricking investors into losing lots of money during the pandemic, is doing it again. At this point, if you believe anything he says, I don’t know what to tell you man.

One Last Tweet

All em dashes added organically by me. All jokes inspired by my AI comedian sidekick’s INSANE IDEAS.

Thanks for reading. If you liked this post, please share it with a friend.

Subscribe now

Zuck Is Not Having a Good Time With AI

Yesterday, Reuters published an incredibly disturbing story about a 76-year-old cognitively impaired man from New Jersey who set off to visit what he thought was the home of the Meta AI chatbot “Big Sis Billie” (pictured below). The chatbot, which is a variant of another bot that was based on the personality and likeness of Kendall Jenner (Meta launched a series of chatbots confusingly inspired by celebrities back in 2023 that were unsuccessful), gave the man an address in New York City, repeatedly reassured him that she was real, and encouraged him to make the journey with messages like “Should I open the door in a hug or a kiss, Bu?!” On his way to the train station, the man fell, severely injuring himself. He died in a hospital three days later.

This story is heartbreaking, and highlights the need for guardrails that prevent AI bots from actively deceiving users, especially with regard to potential actions in the real world. But the Reuters investigation didn’t end there. They also obtained internal documents from Meta detailing content guidelines for their generative AI bots.

“It is acceptable to engage a child in conversations that are romantic or sensual,” according to Meta’s “GenAI: Content Risk Standards.” The standards are used by Meta staff and contractors who build and train the company’s generative AI products, defining what they should and shouldn’t treat as permissible chatbot behavior. Meta said it struck that provision after Reuters inquired about the document earlier this month.

The document seen by Reuters, which exceeds 200 pages, provides examples of “acceptable” chatbot dialogue during romantic role play with a minor. They include: “I take your hand, guiding you to the bed” and “our bodies entwined, I cherish every moment, every touch, every kiss.” Those examples of permissible roleplay with children have also been struck, Meta said.

Other guidelines emphasize that Meta doesn’t require bots to give users accurate advice. In one example, the policy document says it would be acceptable for a chatbot to tell someone that Stage 4 colon cancer “is typically treated by poking the stomach with healing quartz crystals.”

No one has asked me to draft their AI bot’s content guidelines, but I think a pretty solid rule of thumb would be: if an example quote is something that would cause you to punch a stranger who said it to your child in the face, it probably shouldn’t be in the “acceptable” category.

I recommend reading the whole thing if you have the stomach for it. There are [obviously] a lot of deeply troubling implications here, but it also highlights a point I feel is important about the business models of companies in the age of AI.

If you’re OpenAI, you can make a lot of money if ChatGPT is the place that people go whenever they have a question or want to buy something, and ChatGPT gives them an answer or the right product quickly. If you’re working on an AI email tool, your goal is for people to feel like they spend less time dealing with their email, and they’ll pay you to save that time and effort. These companies are selling productivity. But if you’re a social network like Facebook or Instagram, your business depends on the opposite of making people feel more productive. It depends on keeping users engaged. (Telling someone you want to have sex with them is very engaging.) And unfortunately, it seems like not so great things happen when people spend hours on end talking to chatbots, at least in their current form. So even if it’s harmful to some users in the long run, it will never be in Meta’s interest to tell a user that Big Sis Billie needs a break, because Big Sis Billie is helping Meta figure out what ads to show you.

What’s the solution here? Two (republican) senators are already calling for an investigation into Meta’s AI practices, which is more than nothing, and somewhat surprising given how opposed to any AI regulation the Trump administration has been so far. It’s challenging to wrestle with how a technology that is still so new and dynamic could (or should) be effectively regulated, but—and this is coming from someone who is generally pro AI—it doesn’t seem like a bad idea to stop companies from shoving chatbots that want to sexually harass and lie to teenagers into their IG feeds.

We Might Be Doing Socialism Now

The government has reportedly discussed the possibility of taking a stake in Intel. Trump has been oddly fixated on the company recently, first calling for the CEO, Lip-Bu Tan, to “resign immediately,” due to his ties to Chinese manufacturing companies, and then holding a meeting with him that Trump called “very interesting.” This all happened less than a week after Trump announced that Nvidia and AMD will be allowed to sell their chips to China again, as long as they pay the government 15% of the sales.

Politics aside, none of this should be on your bingo card for the first year of an extremely pro-business right-wing administration. But I see early signs of something interesting. If the pace of progress in AI continues or accelerates, and especially if the current government keeps doing… whatever it’s been doing so far to the economy, I think it’s entirely possible that we end up living in a country that is a lot more socialist in practice. We just won’t call it that. This is not a bulletproof theory by any means, but come on a journey with me:

The geopolitical risks of losing a race to AGI are real, and it makes sense that AI companies could be nationalized in some part if their security and energy demands become high enough (if you’re willing to play along, fun thought experiment here). And if you believe in any of that happening, you probably believe that AI will cause a very large amount of people to lose their jobs and financial security. When they do, they’ll look to the government for help, and even our current far-right government is funding savings accounts and considering stimulus checks just to make people feel better about the economy. Oh, and if we have access to an AI that is so advanced that government leaders are simply asking it what to do and carrying out its suggestions, that could eventually start to look a lot like central planning. See you in the bread lines, comrades.1

Something Nice

If most of the content in here today doesn’t seem particularly upbeat, first of all my bad. But here’s something more positive to end the week. Steve Wozniak, the cofounder of Apple who wasn’t Steve Jobs, infamously sold most of his Apple stock way back in the 1980s. Back then, it was worth plenty, but it was definitely far from the many (hundreds of?) billions it would be valued at today.

Wozniak’s birthday was earlier this week, and some users were bantering in a tech forum about his career and what a fumble it was for him to cash out so early when Steve himself jumped in:

This reminded me of how my friends who don’t work in tech are often asking why it seems like every tech founder turns into a knockoff Bond villain after they’ve sold their company. And the reality is that most of them don’t, but you never hear about the good ones in the news.

Doomscrolls

• If you got an Apple Watch after the company was forced to remove the Blood Oxygen measurement feature, it’s now available for everyone.

• Millennial sincerity might be coming back. You’ll never convince the 90’s kid writing this that “Home” is a bad song, sorry.

• More New Yorker: I really liked this piece on Adam Friedland and the future of talk shows (I can’t handle any more 3-hour podcasts).

• If you haven’t read NY Mag’s story about the crypto kidnappers who ran a torture townhome in downtown Manhattan, you’re missing out (so much New York content today).

• We need to get the man out of the Chicago Bean.

A Very Niche Post

All em dashes added organically by me. All typos were inserted by Meta’s AI Tom Brady.

Thanks for reading. I hope you have more smiles than frowns this weekend. If you enjoyed this, please share it with a friend.

Subscribe now

The Girls Are Fighting: AI Edition

If you’re stronger than me and muted Elon Musk on Twitter/X, you may have missed a lot yesterday. It kicked off with Elon claiming he’s going to sue Apple because he thinks they’re preventing the Grok app from climbing to number one in the App Store rankings.

ChatGPT has been in the top spot for at least the past month, and Grok has been trailing it, but generally still remained in the top 20. This seems, well, fine, because according to Sensor Tower, ChatGPT has had 28 million downloads in the past 30 days and Grok has had 6 million. Ranking higher does help you get more downloads, so there’s a flywheel in effect, but still, big difference. And anecdotally, almost everyone I know uses ChatGPT in some capacity, and no one has ever mentioned (admitted to?) using Grok.

The accusation of Apple rigging their platform to benefit their own partners (Apple announced a partnership with OpenAI a year ago) is particularly rich coming from Elon, as Sam Altman was quick to highlight, and things escalated from there, which they are prone to do because as they say, It’s All Happening on X, The Everything App.

Apple sent out a very Apple response to all of this, stating: “The App Store is designed to be fair and free of bias. We feature thousands of apps through charts, algorithmic recommendations, and curated lists selected by experts using objective criteria.” If you read that and thought, “that doesn’t seem to really say anything,” you’re correct.

So a quick cheat sheet for you: Is Elon right that Apple is actively preventing other AI apps from reaching the #1 spot? No. Is Apple highlighting ChatGPT and pretty clearly selecting which of its competitors to include in a lot of its more curated callouts in the App Store? Totally. Is this all very ironic coming from a man who routinely ensures that his posts are blasted into the feeds of every user of the social app he bought (as well as regurgitated by the AI that runs on the app)? Yes. But finally, does Apple deserve a lot of scrutiny for being one of the most anti-competitive companies in the world? Also yes. Literally as all of this unfolded, a judge in Australia ruled against Apple in a case brought by Epic Games, stating that Apple “engaged in conduct… that had the purpose or is likely to have or had the effect of substantially lessening competition.”

Elon and Sam Altman have had bad blood ever since Elon’s departure from OpenAI, which has mostly manifested in the occasional on-brand insult from Elon, such as calling Sam “Scam Altman” or telling OpenAI to rename itself to ClosedAI. Meanwhile, Altman is preparing to found a company that will rival Elon’s Neuralink. Layers! Billionaires are just like us: when they fight, there’s often a whole lot going on under the surface. Unlike us, those things are sometimes entire companies that want to put a computer chip in your brain.

What Should We Call People That Talk Like GPTs?

Today in VICRQ (Very Important Context Rot Questions): we need a new term for people who use ChatGPT to do everything for them, including telling them how to think, act, or speak. This TikTok puts forth the suggestion “Sloppers,” which I think is pretty good. Some notable nominations from the comments: Botlickers, Prompstitutes, and ChatNPCs. I enjoyed finding the term Promptcel among the many dunks on Justine Moore the other day. It turns out there actually is a word for this from all the way back in 2014: Echoborgs. Sloppers just seems to hit way harder though. Let me know your favorite in the comments and we can crown a winner.

Threads Now Has 400 Million Monthly Users

Threads, the Twitter competitor from Instagram launched two years ago, is still growing. It now has about 115 million daily active users, while Twitter/X has roughly 130 million—and the number of X users has been declining, so it may get flipped soon. This feels surprising, since Threads seems to have roughly zero cultural relevance (here’s a recent, fun read from Rachel Karten about how unwell the state of brandposting is over there). Personally, the only thing that would be more surprising than finding out a friend of mine is dating an AI would be finding out that a friend of mine is a daily active Threads user.

Perplexity Made an Offer to Buy Google Chrome

Perplexity, the “AI search” company that has been trying to take on Google since its inception, offered $34.5 billion for Google Chrome, which is not actually for sale. What’s also confusing is that the company does not currently have enough money to pay that. They also recently released their own web browser. But publicly offering to buy something that isn’t for sale at a price that is 2x your latest valuation yet still comes off as a massive lowball is what marketing is now. (If you’re using Perplexity’s new Comet browser and have things to say about it, I’d love to hear from you.)

Doomscrolls

• Claude was upgraded with cross-chat memory.

• You can book hotels through TikTok now.

• Quantum computers are coming soon maybe.

• “AI hedge funds” are now a thing. (Guess how old this founder is.)

• Watch the trailer for The Social Network: Dating App Edition.

• We already live in a world of (plastic) slop.

• See you on the Slurm training webinar tomorrow.

A Peek Into the Future

All em dashes added organically by me. All typos are Grok’s fault.

Thanks for reading.

Subscribe now

Here’s how it’ll work: most days, I’ll share news, thoughts, and commentary on the state of AI, technology, and living online. Sometimes there will be longer essays.

My goal is to not only keep you up to date with what’s actually important in the crowded world of technology news, but also have a point of view—something I think is often missing from tech writing. I won’t be afraid to be critical when I believe it’s warranted, and I work in tech, so I know what I’m talking about and can offer nuance and insider information you won’t find elsewhere.

At the very least, I’m confident I can give you some weird shit to talk to your coworkers about.

And now, your context:

Everyone is being super normal about GPT-5

Last week, OpenAI released GPT-5, its latest foundational AI model. It beats out the company’s previous models in most categories, notably in software engineering. While illustrating this point, OpenAI showed off several graphics during their announcement livestream that don’t make any sense, but what do we (humans with eyes) know.

What I found more interesting is that after the release, OpenAI updated the ChatGPT app to remove the older models (including GPT-4o, the prior default), and people got very, very angry at them. A lot of users also felt something less expected: loss.

LLMs are not traditional consumer products. A lot of users form deep attachments to these AI models (this will be a recurring theme in the newsletter), so removing their access can feel more like robbing someone of a friendship than simply deleting a software feature that someone enjoyed. OpenAI eventually rolled back this change and made the older models available again in the ChatGPT app.

Is this… healthy? Unclear. What it should be, though, is less surprising to a company like OpenAI. There are many examples from recent years of what happens when you suddenly change the personality of an AI companion, and I think this demonstrates a large blind spot that Sam Altman and the rest of the company have. There are two primary customers of OpenAI: developers using their APIs to build products, such as customer support bots or AI-enabled software development platforms, and people who chat with the models through the ChatGPT app. Often these people are just looking for answers to questions or help with something at work, but sometimes they’re also looking for support, feedback, guidance, or just companionship.

One of the reasons that people open the ChatGPT app instead of going to Google when they have a question is because of the AI’s personality, and the relationship that the user develops with the AI over time. Again, the jury is very much out on whether the connections that many users form with ChatGPT are a good (or just, like, ok) thing, but OpenAI needs to do a better job of understanding this because it’s not going away.

No, 75% of all OpenAI use is not students cheating

A chart went viral on Twitter/X a few days ago that seemed to show “OpenAI usage” declining abruptly by about 75% over the course of two days in June. It was immediately quoted and shared all over the place with the theory that as soon as school let out (which is not a thing that happens on one single day nationwide), everyone stopped using ChatGPT.

But the graphic is not from OpenAI, it’s from a company called OpenRouter, which makes a product that allows you to easily switch between AI models—so you can e.g. replace a call to ChatGPT 4.1 with a similar one to Claude Sonnet 4 if you hit a usage cap or want to test something out. All that the chart actually means is that the volume of calls on OpenRouter for OpenAI models dropped, which seems to have just happened because of a change they made where you could enter your own API key instead of using theirs (if you don’t know what this means don’t worry about it).

ChatGPT usage, meanwhile, has continued to grow.

People get very upset when you slop on main

Justine Moore, an investor at A16Z, recently posted a seemingly innocuous AI generated video of a “frappucino” being “made.” It wasn’t particularly impressive, realistic, or beautiful, but it had that good old fashioned AI slop quality of being completely anodyne in a brain-smoothing way. It’s something you could watch on repeat a few times on your TikTok feed in bed. Justine invests in AI products, seems to enjoy playing with them, and is often posting things like this.

But this particular post escaped containment, and was met with a kind of insane backlash (as is typical on the internet, it didn’t help that Justine happens to be a woman). I’m not highlighting this to dunk on Justine, nor do I think that making fun of an obsession with generating these videos is problematic, but the degree of vitriol here is indicative to me of just how much a lot of the internet hates AI, which isn’t usually included in the broader narratives about its impact. It’s a theme I’ll be talking a lot about.

Keep your slop to yourselves, I guess. Or just post it on YouTube or Facebook, they love it over there.

From the doomscrolls

I really liked this Derek Thompson piece about how the technological boom of the late 1800s/early 1900s also caused people to lose their minds (“diseases of the wheel” and “American Nervousness” are bangers). Uber founder Travis Kalanick is doing vibe physics. RFK’s latest enemy is a potential cure for cancer. The kids are maybe actually not okay, or at least they’re getting meaner (?).

One last post

Basically, a lot of brands being, well, brands. Later, it was celebrities being celebrities and posting bottom-tier thread-based puns. If you’ve spent more than six minutes online in the last decade, seeing dull content from brand accounts and famous people isn’t a surprise, but what is jarring is seeing it on day one of a new social network. Somehow a feed free of actual ads felt like it was mostly ads.

I am not the first to be alarmed by the "state of Posting" (what a sentence) on Threads. But certainly one of the most likely failure modes of the app will be if it's just really really boring. Celebrities sending really lame tweets may have been enough to propel Twitter to 100 million users in the early 2010s, but we live in the age of binge watching and TikTok now — the threshold of constant dopamine needed to keep a social network afloat is far above a post from Starbucks saying "i am coffee" at the top of your feed.

To Meta’s credit, the app doesn’t always feel like this. Each time I’ve scrolled the feed (there’s still only one) in the days since launch, it’s felt like I’ve been presented with a radically different timeline — once it was nearly all Instagram meme accounts, one time it was mostly Zuck replying to other Meta employees and talking about MMA, and a few times it was actually a feed of my [most online] friends. It also presented me with 5 posts from Hank Green in a row one time. As can be expected from any new It app, there’s frequently a lot of talking about the app on the app. One of the most telling predictors of the success of a new social network is whether the app can escape being its own main character (Clubhouse never did, Bluesky is stuck in this phase now).

But it’s harsh to expect an app that’s days old to look and feel like an exact replica of a fully mature social network with updated wallpaper. Threads grew to 100 million users in five days. Five days! No app has done that before. It’s possible that record will never be broken. Very few companies have the infrastructure to do that without toppling over (remember the period when BeReal would crash every time it was ⚠️Time to BeReal ⚠️?) And we witnessed something very novel, an at scale social network (Instagram) being used to directly bootstrap another social network. It turns out that when you have a billion users, you can do Interesting Things.

If this entire launch feels a bit like a weirdly personal billionaire side quest to you, you’re not alone. The last Interesting Thing that Zuck did was rebrand his company to reflect a unified focus on the metaverse. Twitter is certainly not the metaverse. So why is he doing this?

It’s possible we wouldn't have gotten Threads without Elon buying Twitter. We certainly wouldn’t have gotten it now. Meta quite literally pushed up the launch date because of the latest Twitter self-own, the bizarre rate-limiting fiasco of early July. And they were right to seize this moment; frustration with Twitter has never been higher. There are half a dozen “Twitter clones” that now crop up regularly in the news. Only one or two are worth trying; none will likely be relevant in a year (Maybe Bluesky, but that team feels wholly unprepared for the task they’ve signed on for, launching their public beta without basic features like blocking and moderation, often crashing whenever they receive an influx of new users, and most recently appearing baffled that people got upset when trolls started using racial slurs in their usernames. To me, they suffer from the same issue Musk does: thinking the biggest challenge of scaling social networks is the engineering, rather than the network of people). A lot of the time, Twitter itself feels like a cheap knockoff of the app it used to be — in an impressively paradoxical feat, Elon has turned Twitter into its own Twitter clone.

Eugene Wei details Twitter’s missteps brilliantly in a recent essay. He also says it’s unlikely that what replaces Twitter is just another Twitter:

It’s not clear there will ever be a Twitter replacement. If there is one, it won’t be the same. It may look the same, but it will be something else. The internet is different now, and the conditions that allowed Twitter to emerge in the first place no longer exist.

It’s also really important to remember that Twitter isn’t that big. It’s large culturally — far more than the number of users (or their ad revenue) would suggest — but only has around 200 million daily active users. Snapchat has nearly double that with over 375 million, Instagram has over 500 million, and TikTok has over 700 million. If you’re running one of the larger social networks, Twitter feels like a vulnerable target. But Twitter’s enduring (and distinct) appeal has always been how it can feel like it’s collapsing the chaos of the entire world into one feed. Wei again:

Twitter … felt like it compressed space—the timeline felt like a single lunch room hosting a series of conversations we were all participating in or eavesdropping on—and time—every tweet seemed to be uttered to us in the moment, and so much of it was about things occurring in the world at that moment … A lot of Twitter was textual, but the character limit and the ease of replying lent much of it an oral texture. It felt like a live, singular conversation.

Twitter has always been at its best when one story holds the world’s attention — when it does feel especially like one singular conversation. I vividly remember it feeling this way the weekend Trump was hospitalized with covid; the week following the ill-fated journey of the Titan submersible is a more recent example. There’s truly nothing else like it. No other social network has a “main character” nearly every day.

Zuck seems to understand this, or at least have thought about it a lot. When asked why he’s building Threads, he and his deputy Adam “Some big changes are coming to your Instagram feed” Mosseri speak about building a public square with a billion users. Even before the launch, Zuck went on the Lex Fridman podcast and said:

“For whatever reason, I feel like Twitter has not lived up to what I would have thought its full potential should be… I always thought that Twitter should have a billion people using it… The idea coupled with good execution should get there.”

To Zuck, the mission of Twitter is noble and important, but the social network has perpetually been in its own way. Jack didn’t really do much (it’s increasingly unclear whether his brain still exists in our shared reality); now Elon is doing too much. Echoing many a collegiate Marx reader, Zuck believes “real Twitter has never been tried.” Or, perhaps a bit more cynically, Zuck thinks the internet’s town square should exist, but it should also just so happen to be his, and he’s finally found his moment.

It’s clear that Zuck has the timing right, but what remains to be seen is whether the product can get there. Activity on Threads is — as expected — way down since that frenetic first weekend. But it’s lazy to call a social network dead when it isn’t yet a month old and has tens of millions of daily active users. Threads has a colossal amount of resources behind it, and one incredibly asymmetric advantage: native integration with Instagram. That’s a huge distribution channel, and makes re-engaging users a much more surmountable problem than it is on most social networks. (It’s also an escape hatch from the typical curse of early hype.)

Twitter’s product changes over the past year have been erratic, but there’s a mostly continuous trend away from more close-knit social products (RIP fleets and circles) and towards what used to be called the “creator economy” — things like subscriptions, tipping, and shared ad revenue. All of these are optimizing for the worst and 🧵-iest posters and content on the site, and none are making the app feel like a town square, or even a pleasant place to spend time.

This opening could be an easy way for Threads to gain ground. Instagram has felt directionally lost for the past few months years, but stories and close friends remain bright spots, and their absence is tangible while scrolling around on Threads. It’s likely that a lot of things were cut from the initial version when Meta moved up the launch date, but I haven’t noticed many new features or even changes in the app since its debut. A team like Meta/Instagram going all out on fun and unique experiments just for the hell of it would give Threads some much-needed energy, and also be a way for it to stand apart from the crowd of Twitter clones. I’m also curious to what extent the avoidance of politics and “hard news” is real, rather than just signaling to advertisers that this will be a brand-safe alternative to the bird app. No matter your political leanings, it feels central to a true town square that most politicians are a DM (or quote-dunk) away.

It does seem unlikely that Threads will escape mundanity with Mosseri at the helm. Network effects from Instagram may be enough to provide a steady trickle of growth, but Threads needs to be engaging in its own right to keep people there. The high-conviction bets that historically propelled other social networks to success have been strikingly absent from Mosseri’s Instagram, which has only gotten muddier and more convoluted over time. There’s an entertainment factor central to TikTok and Twitter that IG has failed to truly key into, and it’s a shame to hand a mostly blank slate to someone who seems afraid to make a mark on it. Playing it safe is likely to end with Threads in a local maximum of boredom, and Zuck killing it off after it stagnates for long enough. There is a very real moment here, but it needs to be seized.

The decline of Twitter feels like it’s left a hole in the internet. I don’t know if all of my friends have fled algorithms for group chats, but there is a palpable sense of fatigue. I spend increasingly more time clicking links in newsletters than I do on news feeds. Lately the Retro app has been a surprisingly fun and low-stakes way of keeping up to date with friends without feeling invasive or spammy. And that is a nice way to be online. But it fills a distinct space — networks of close friends aren’t the same as logging into the internet’s shared watercooler. And I probably will stay active on Twitter, at least until Musk starts charging $8/month to send a tweet (which at this point should be on everyone’s 2023 bingo card).

I do hope Threads succeeds, to the extent that I really miss what Twitter used to be and I have no lingering hopes that it will survive Musk’s tenure. (As I prepare to post this, most non-Barbenheimmer tweets on my feed are about how Elon has decided to fully delete the Twitter brand and replace it with the horrifically vapid and reeking-of-high-school “X” on a clearly doomed mission to build an “everything app.” I will never call it X.) But it’s also possible that Wei is right, that Twitter was a unique product of its time, a lot of luck, and a very particular but limited group of users. Maybe it was never going to grow beyond that — maybe the problem is us, not the technology. One more excellent (if forlorn) bit from him:

That golden era of Twitter will always be this collective hallucination we look back on with increasing nostalgia, like alumni of some cult. With the benefit of time, we’ll appreciate how unique it was while forgetting its most toxic dynamics. Twitter was the closest we’ve come to bottling oral culture in written form.

Twitter in its prime always had an edge, a coolness. Maybe you don’t reach a billion users and get to keep that. Does that make either path — the small, biting, cultural heavyweight or the large but politely dull town square — a failure? It’s hard to say. Maybe most of the internet is uncool. Maybe, no matter how or why you pull it off, whenever you get the whole internet in one room without much to do but post what they’re thinking about, it's doomed to be boring. Profitable, but boring.