Deep explorer in AI. Head of Product. Use Twitter to manage interesting links / ideas.
Cambridge
Joined March 2010
- Tweets8.7K
- Following2.5K
- Followers547
- Likes1.6K
James Stevens retweeted
The correct answer to this is to lean in: fees and, if you don’t want to ration by price, lotteries.
The status quo where peoples’ time was wasted queuing anyway isn’t worth preserving.
agents are going to cause a breakdown of human behavior assumptions underlying today's systems. i've started thinking about it as friction collapse. without friction, a number of norms will collapse.
consider selling tickets to a concert/event. today, you can post them at a certain time and it's first come first serve. soon everyone's agent will persistently monitor, and as soon as they are posted for sale, thousands of people will instantaneously be trying to buy the same ticket. how do you even resolve this? sell the ticket to the person whose agent gets there fractions of a second before?
we're seeing the same thing with job applications, restaurant reservations, but basically anything that relies on participation requiring effort is going to break down. sales emails/calls, participation in government benefits, apartment rentals, doctors appointments, RFP responses. customer support won't be able to handle the user who will not drop their point of dissatisfaction. with no friction, these systems just don't work anymore.
business models will break down:
- flights/hotels will be consistently cancelled and rebooked at the lowest price (great use case for your agent if you don't have this already set up)
- subscriptions will be cancelled
- retailers won't be able to arbitrage price differences as the agent will find the cheapest place to buy anything
- insurance denials (health/auto/home/etc.) will be fought tooth and nail
- bank interest rates will be optimized and avoidable charges will be avoided
a lot of what we perceive as stable social coordination is actually an equilibrium sustained by human inefficiency. rationing is a product of human slowness. signaling effort/interest/intent is a key part of our social infrastructure. inertia is a key to many business models. information asymmetry is everywhere. and attention is monetized thoroughly.
with agents, none of these things are true. humans will simply demand what they want and have their agents get it for them. there is going to be a fundamental renegotiation of how markets allocate resources, how people signal intent, and how businesses capture value.
James Stevens retweeted
Dusting off this tweet from April because this gap in shared understanding of LLM capability is *widening*. It's now less "two groups of people speaking past each other" and more a sharp funnel.
- Napkin math somewhere around 6B people (~75% of the population) have barely come in contact with LLMs at all.
- Around 1-2B (~20%) are casual and infrequent users of free-tier ChatGPT-like products. This group sees derpy chatbots and treats them a bit like a better Google search, a writing aid, or etc. Maybe an agent tries to book you a flight. I have non-tech friends who (reasonably, imo) say they have not much use for it at all. Even many professionals outside of math&code are in this tier. For example, execs and many other functions spend a lot of their time talking to other people, so while they understand what is happening intellectually, it is still second-hand and a bit abstract.
- Now we get to professional use of frontier-grade LLMs in math&code. Somewhere around 20M people (0.2%) see first-hand that large, complex projects that used to take them weeks/months can now be completed by agents with a prompt. Building apps, copying apps, translating apps, decompiling apps from binaries... This has all happened very quickly and recently - less than 1 year ago, I was writing code manually by hand, typing memorized computer code commands into a code editor character by character, occasionally pressing Tab to autocomplete a little chunk of code.
- And finally we get to the ~5,000 people (~0.00006%) with access to frontier-grade systems internally. The external world has seen the preview. It looks like swarms of thousands of agents collaborating over weeks on software mega projects: minting zero days, running cyber attacks and defenses at machine speeds, discovering new science, advancing the frontier of mathematics. Things that would have taken top professionals in the industry years of work. Meanwhile, human review and comprehension are starting to fall behind. For example, people are still involved in the "archeology" of the OpenAI-HF incident from many months ago. Mathematicians may be poring over the 722 manuscripts on frontier mathematics for a while.
The funnel is driven by a combination of factors:
1. The impact scales with ambition, problem size, and horizon. A question with a paragraph answer barely stresses the system. You need a reservoir of big, difficult problems that you really care about. This aspect drives the consumer / professional dimension of the funnel.
2. The jaggedness of the system (which I have written about a lot separately). Capability peaks in domains that are digital, verifiable and economically valuable. This is because LLM capability emerges from reinforcement learning on verifiable rewards on a curated environment mixture driven by revenue potential. This aspect primarily drives the area (e.g. math&code) dimension of the funnel.
3. Access. Free-tier, paid-tier, internal.
So this is the weirdness of the moment. The general public has mostly not interacted with these systems. When they have, it looks like a derpy chatbot. The majority of professionals still see only a modest uplift. And a small sliver of professionals are experiencing the vertigo of the curve going vertical. And it is all happening at the same time.
Judging by my tl there is a growing gap in understanding of AI capability.
The first issue I think is around recency and tier of use. I think a lot of people tried the free tier of ChatGPT somewhere last year and allowed it to inform their views on AI a little too much. This is a group of reactions laughing at various quirks of the models, hallucinations, etc. Yes I also saw the viral videos of OpenAI's Advanced Voice mode fumbling simple queries like "should I drive or walk to the carwash". The thing is that these free and old/deprecated models don't reflect the capability in the latest round of state of the art agentic models of this year, especially OpenAI Codex and Claude Code.
But that brings me to the second issue. Even if people paid $200/month to use the state of the art models, a lot of the capabilities are relatively "peaky" in highly technical areas. Typical queries around search, writing, advice, etc. are *not* the domain that has made the most noticeable and dramatic strides in capability. Partly, this is due to the technical details of reinforcement learning and its use of verifiable rewards. But partly, it's also because these use cases are not sufficiently prioritized by the companies in their hillclimbing because they don't lead to as much $$$ value. The goldmines are elsewhere, and the focus comes along.
So that brings me to the second group of people, who *both* 1) pay for and use the state of the art frontier agentic models (OpenAI Codex / Claude Code) and 2) do so professionally in technical domains like programming, math and research. This group of people is subject to the highest amount of "AI Psychosis" because the recent improvements in these domains as of this year have been nothing short of staggering. When you hand a computer terminal to one of these models, you can now watch them melt programming problems that you'd normally expect to take days/weeks of work. It's this second group of people that assigns a much greater gravity to the capabilities, their slope, and various cyber-related repercussions.
TLDR the people in these two groups are speaking past each other. It really is simultaneously the case that OpenAI's free and I think slightly orphaned (?) "Advanced Voice Mode" will fumble the dumbest questions in your Instagram's reels and *at the same time*, OpenAI's highest-tier and paid Codex model will go off for 1 hour to coherently restructure an entire code base, or find and exploit vulnerabilities in computer systems. This part really works and has made dramatic strides because 2 properties: 1) these domains offer explicit reward functions that are verifiable meaning they are easily amenable to reinforcement learning training (e.g. unit tests passed yes or no, in contrast to writing, which is much harder to explicitly judge), but also 2) they are a lot more valuable in b2b settings, meaning that the biggest fraction of the team is focused on improving them. So here we are.
James Stevens retweeted
Sigh.
I wrote this post below, but then I didn't post it because I am tired of the consistent pushback I get from folks with the same safety worldview as Anthropic. I guess that means I should send it. Here goes!
This post is pretty solipsistic and doesn't properly take recent events in cyber risks into it's discussion. It is really hard for me to watch how the US frontier labs like Anthropic don't have the ability to consider other approaches to safety and ways things could play out.
It insinuates that Z ai (Chinese lab, builds GLM series) doesn't really care about safety and is reckless to take their business strategy.
Saying things like "Given this evidence, we think it's likely both state and non-state actors will use models like GLM-5.3 to cause real-world harm." and "This is unlike any other similarly capable AI model, all of which were released with safeguards or through limited access programs." while closed models have been used on more of the documented cyber attacks is just bowing out of the interesting question.
A plausible view is that open model weights and closed model apis (with some safe guards) are both far closer to being easy to mis-use, rather than API models being closer to safe. The trope "Open Dangerous, Closed Safe" may be closer to "Open Unsafe, Closed Unsafe"
Closed models have stronger capabilities and stronger safeguards, but the stronger capabilities part could matter more in net harm if both the safeguards are porous.
At the same time, as the authors do acknowledge (thanks - thats progress!), open models without extreme cyber guardrails are important to rapidly diffuse cyber readiness in the economy -- as programs like project glasswing are not perfect in getting all critical industry onboarded.
There will be more issues like the time when Fable was released and Amazon found a workaround, which allowed them to access the full capabilities of the model served readily at an API.
The blog overall is reasonable in it's narrow line, but it's a very effective tool in a complicated, rapidly evolving media ecosystem to reinforce a certain type of safety thinking.
James Stevens retweeted
big AI news
Google just demonstrated a recursive self improvement loop for AI discovery
Google/DeepMind researchers introduced Dream-RSI, a system where an AI agent improves how it explores problems by replaying its past discovery attempts, testing thousands of alternative strategies cheaply, then deploying the better strategy in the next round.
Across algorithm design, mathematical optimization, and GPU kernel engineering, it matched or improved discovery quality while cutting search costs dramatically, in one setting reducing agent calls by up to 162x. 👀
Importantly, it improves the exploration policy, not the underlying model weights.
James Stevens retweeted
By now, it's amply clear that the big bosses of US frontier AI labs are really, really scared of China.
And, fascinatingly, part of the answer why might be a tiny place in Inner Mongolia that you've probably never heard of: Ulanqab prefecture.
Heard of it? Didn't think so.
Even myself, who's traveled to Inner Mongolia twice, had never heard of it before researching this story.
This is because, at first glance, this place is pretty unremarkable. Check its Wikipedia page (en.wikipedia.org/wiki/Ulanqa…) and you’ll see the only thing it’s famous for is being the birthplace of He Pingping, who briefly held the Guinness record for world’s shortest man.
That’s it, that’s the list!
Otherwise, it's just a sparsely populated - 1.5 million people, minuscule by Chinese standards - stretch of windswept grassland on the Mongolian plateau.
Except there is one number that tells you there is more to this place that meets the eye: this tiny prefecture consumes nearly 1% of all of China's electricity - and that number is growing by double digit numbers every year.
And yes, I did write **consume**, not produce.
In fact, when you divide Ulanqab's electricity consumption by the number of households living there, each household in this remote corner of Inner Mongolia "consumes" about 105,000 kWh in electricity, roughly 10 times the U.S. average. Insane electricity consumption numbers.
So what are the Ulanqabese doing? Are they all running 10 American households' worth of appliances?
Here is another number that might clue you in on the answer: over the past few years, this unremarkable prefecture has quietly signed over half a trillion yuan (and, yes, that’s trillion, with a "T") worth of investment deals from all the major Chinese tech giants.
What is going on, you'll have guessed by now, is the most ambitious data center buildup anywhere on earth - and it's really surprising this hasn't been talked about more because the scale is beyond anything else, and by an immense margin.
Take Elon Musk's so-called "Colossus" datacenter in Memphis, Tennessee which he pitches as "the world’s largest AI supercomputer."
According to their own numbers (x.ai/colossus), Colossus has 200,000 chips, which, let's be clear, is already super impressive.
In datacenter lingo, this converts to roughly 5000-6000 "racks": you know, the fridge-sized cabinets full of blinking lights you see in every movie scene set in a server room.
How many racks are they building in Ulanqab? Over 5 million.
Yes, about one thousand times the scale of "Colossus" 🤯
I tripled-checked the number: it is indeed the official number published by authoritative sources such as "Science and Technology Daily", the official newspaper of China's Ministry of Science and Technology (stdaily.com/web/gdxw/2026-08…).
This is an absurdly large amount of compute.
What we see appearing in this Inner Mongolian steppe may be the closest thing to a world brain humanity has ever built - a place where a large share of the world's thinking will physically happen.
And - as we'll see - what makes this story so fascinating is not just the scale, but why it's getting built in this particular patch of Inner Mongolian grassland which combines a number of characteristics unlike anywhere else.
If you believe, as I do - and as both China and the US obviously do as well - that AI will be the defining technology of the 21st century, this makes Ulanqab one of the single most relevant geopolitical places in the world right now.
The full story - the Chinese government plan Ulanqab is part of, energy prices, chips and the strategic questions this raises - is here: arnaudbertrand.substack.com/…
Be careful out there all you AI post generators.
Algo gonna get you.
NOTE: 💯% not denigrating @ttorres who's awesome and doesn't do slop! I was sharing this post when I noticed it.
James Stevens retweeted
Massive release from Perplexity this morning
Big win for local ai, possibly a real tipping point 🔥
In a compute and power-constrained world, a good chunk of agentic inference needs to move to local hardware. A drastic version of that is a fully local agent runtime, where the model (orchestrator and subagents) and the harness run locally. Portable Computer from Perplexity is this. Launching today for @nvidia DGX Spark.
What if all the AI Vampire exhaustion, need to simplify Claude outputs (teapawt.net/mykola/status/20814504…) using things like ASD-STE100 and intellectual jargon has all been caused by watermarking?teapawt.net/pbaylies/status/208893…
Went through the Anthropic posts, watermarking won't swap potential words for stranger ones. Model still picks the shortlist of words that fit and watermark only decides which of the equally good ones wins. Change of coin flip not judgement.
