John Croxton (@SterlingCroxton) profile picture

Tarbell fellow writing about AI. Tech, defense, innovation, bureaucracy and rough numbers.

Joined July 2017
Photo by John Croxton (@SterlingCroxton)Photo by John Croxton (@SterlingCroxton)Photo by John Croxton (@SterlingCroxton)Photo by John Croxton (@SterlingCroxton)Photo by John Croxton (@SterlingCroxton)Photo by John Croxton (@SterlingCroxton)Photo by John Croxton (@SterlingCroxton)Photo by John Croxton (@SterlingCroxton)Photo by John Croxton (@SterlingCroxton)
News that the Riemann Hypothesis, the Holy Grail of math, has been solved by Claude. The obvious question for AI is: what's the equivalent of the RH in the field of AI research itself? We could see extremely fast progress soon.
My latest story for @DefenseOne, on AI automating its own research and policy moving slowly: Feat. @DKokotajlo, Joe O'Brien, and Frank Cilluffo
1
2
44
Also note that the Riemann solution hasn't been confirmed, though it seems more likely correct than not
1
9
METR's report on R&D automation Thursday was a huge shock to the team behind AI 2027, and probably the single biggest news in AI this year. AI takeoff in 2027 looks pretty likely, unless there's some kind of imposed slowdown. Much faster than many people (including me) expected.
Unfortunately, it seems like the intelligence explosion will be happening in 2027 after all. Most of this post analyzes the new METR report (though other sources of evidence are also important). If they’re right, that implies an AI-driven acceleration that is 5 months ahead of the AI 2027 scenario. I am scared, sad, and angry. blog.aifutures.org/p/q35-tim…
2
2
15
1,167
Hard to wrap your head around: 1. Anth spending ~75% of compute on R&D 2. Them increasing revenue 10x YoY but only ~3xing total compute Did they add more than 3x more compute YoY? Did they *decrease* their fraction of compute to R&D from an even higher % last year?
If I'm not mistaken, Anthropic's 85% gross margin and >40% compute margin from this Bloomberg story implies that 70-75% of their compute is now going to R&D. bloomberg.com/news/articles/…
1
1
273
Maybe some of the numbers reported have gotten mixed up. It’d also be worrying if Anth recently increased their share of compute for R&D - that’s what you would expect if they were starting RSI.
30
Update: I might be wrong about this! Waiting for an ARR number to confirm, but my current read is that Anthropic revenue growth has started to slow from its insane 10x/year pace.
I’m expecting Anthropic to report ARR around $100B tomorrow, more likely over than under. That’s more revenue than Disney. Anthropic is five years old. Get your advance market insights here: joansam.github.io/metr-chart…
77
I’m expecting Anthropic to report ARR around $100B tomorrow, more likely over than under. That’s more revenue than Disney. Anthropic is five years old. Get your advance market insights here: joansam.github.io/metr-chart…
1
1
429
This seems right - strong math results make it more likely that we'll see major theoretical AI breakthroughs. Timelines for this look totally unclear to me. Also unclear what this means for regular day-to-day engineering work on the next-gen versions of today's systems.
The math breakthroughs OpenAI released are (while indirect) a positive update for strong RSI. Strong RSI is most likely to occur if AI can make conceptual breakthroughs in machine learning theory on par with, or even larger in impact than, the development of the transformer architecture. We haven't seen AI produce anything like that in machine learning so far. But math and ML share many similarities as domains. (You could even call ML a branch of mathematics.) Both are highly verifiable purely in compute and rely on symbolic manipulation. AI making true breakthroughs in math (category D in Abhishek's taxonomy below) should lift expectations that it can make conceptual breakthroughs in machine learning as well, which is bullish for RSI. I remain convinced that we don't have evidence of anything even remotely close to a self-sustaining RSI loop today. But if the ability of AI to conduct ML research advances as quickly as it recently has in formal math, that may well change.
3
250
This is measured by their score on Text Arena, an eval which is entering its third year of obsolescence
BREAKING: Google overtakes Anthropic in the odds to lead the AI race at year-end, worrying investors ahead of Anthropic’s IPO. 43% chance.
1
36
I wonder if this is based on OpenRouter data, where OpenAI discounted its models to half price exactly so it would look better on these sorts of watchlists
If this data from @tickerplus is even broadly correct, isn't this really, really bad?
Readers added context they thought people might want to know
Chart bars are equally spaced despite unequal intervals (months left, weeks right), distorting growth visuals and plateau. TickerTrends estimates $76.8B ARR Sep 24; Anthropic last reported $65B end-July. Other mid-Sep reports cited >$100B pace. museum-chart-crimes.vercel.app/posts/anthropi… blog.tickertrends.io/p/openai-and-a… cnbc.com/2026/08/17/ant… axios.com/2026/09/18/ant…
1
2
136
Big announcement from OpenAI. Afaik this is the second(!) instance of OpenAI scrapping a model due ~only to safety concerns. If true, this should put the case that all the safety stuff is regulatory capture to rest - will be terrible for OAI finances, and they want to IPO soon
2
392
OpenAI is just offering 50% discounts *on OpenRouter specifically* because people use it to try and read industry trends. Pretty sneaky move.
Extraordinary share gains for OpenAI vs. Anthropic over the last two months. Per Openrouter, OpenAI has gone from 20% share to 50% share vs. Anthropic (meaning Anthropic has gone from 80% to 50%) since June.
3
288
OpenAI's making the same commitment! Big day for 3rd-party oversight...
I agree with Dario that we need to pace the frontier. This has been a primary topic of discussions we've had at OpenAI in recent weeks. Committing to having independent evaluators with employee-like access is a great idea, and we will do the same. We'll have more to share soon.
4
420