Przemek Chojecki | PC
@prz_chojecki
Research-level Math Data + hard STEM & coding RL tasks for LLMs @ PhD in mathematics.
1.1K Following 15.2K Followers
A Severe Misalignment of AI in Mathematics
New blog post by Terry Tao and a new anti-AI declaration signed by 25 Fields medallists.
I'm not sure what the intendent effect should be. The ban of AI use in mathematics and science in general because of "math community"?
Fortunately math community is broader than just math academia, and thanks to AI, booming like never before.
Also this "mass production of true/false statements" is such a wrong description. As if there's no proofs by AI that you can study, deepen your knowledge, build upon, engage. The "AI slop" is still better written than what 90% of mathematicians write.
Mathematics doesn't belong to math academia. It is a tool to understand the world and help other domains with necessary tooling. AI boosts the understanding and engages wider community.
Swan song of academia.
Link:
Show more
data is the moat
This is notable. DeepSeek, a lab usually first to pioneer novel algorithms and architectures, is saying that at this point, the ROI of improving data quality far exceeds that of working on novel post-training algorithms.
I think this has already been true for some time for non-lab practitioners. If you're doing llm post-training, 80% of your effort should go into looking at your data.
This means:
- Hiring experts to dig through your RL tasks
- Sifting through rollouts and sft data by hand to remove suspicious samples. Make sure all tasks are actually passable.
- Making sure your data is diverse in both difficulty and category.
Show more
Even on $200 Pro plan there are limits to Astra Pro now.
To make sure our current users have an incredible experience and continued access to Astra, we are going to pause subscriptions to our $200 Pro plan. These put the most strain on our systems and we wanted to take the smallest step that allows us to continue giving the broadest access possible. All other plans and the api remain available.
There is no impact to existing accounts and we are working on adding more capacity as fast as we can. Thanks!
Show more
Caltech math undergrads have been saved from AI!
Phew, that was close!
Academia status quo continues for another day.
Thank you for sparing math academia the terror of undergrads getting access to the tool that's increasingly essential to modern math research. We can rest assured they'll all spend the next 6 years of their lives writing pencil and proofs and everything will go back to normal.
Show more
Some more thoughts re Navier-Stokes by OpenAI
- Not entirely unexpected as this was supposed to be the Millennium Prize problem the most susceptible to finding a counterexample with known methods, but still, unexpectedly early. I would bet on 2027 for a problem of this calibre to be solved.
- I'm still pretty much shocked by it. Being a mathematician myself, Millennium Prize problems were always a pinnacle. There are of course more encompassing conjectures (Langlands programme), but these are the single most known problems out there.
- Funnily, I'm not sure it's AGI. Or we already had AGI with GPT-5.5 and Fable, because this is more of the same. 10,000 agents, 130b tokens, scaling test-time compute is the way to go, but isn't really novel nor very AGI-like.
- There's a very real cost to this result in millions of dollars of human time-compute. There are not many businesses that could throw compute at single problems: building rockets or drug research or hedge funds perhaps.
- I'm still pretty conflicted about this collective of narrow super-intelligences aka modern LLMs with harnesses. Are there diminishing returns past certain intelligence point? Is it all about the problem choice now to maximize your ROI? There's a reason we see less economic impact of AI than expected from this level of math breakthroughs. Generalization still seems to be the main missing skill that's hard to RL.
Show more
I tried posting about Navier-Stokes solved by OpenAI on Reddit.
The thread was removed within a minute.
They don't want to know!
Still no Navier-Stokes thread on r/math. Be so for real with us mods
Let's do some rumor-driven Millennium Prize problem work with GPT-6 Astra
OpenAI has solved the Navier-Stokes Millenium Prize Problem with their internal AI model.
Controversies and verification aside, this is bigger than anything so far. Remember 12 months ago people were not sure an LLM can compete well on International Math Olympiads. Here we are.
Show more
We’re sharing a solution to the Navier-Stokes Millennium Prize Problem, one of the deepest problems at the frontier of mathematics.
The proof was produced by a group of agents, using an OpenAI next-generation model significantly more capable than GPT-6 Astra.
The problem concerns whether the description of smooth three-dimensional fluid motion modeled by the Navier-Stokes equations can break down. It has remained unresolved for roughly 90 years.
Show more
The Growing Map of Open Mathematical Problems.
We mapped 15,000+ conjectures from UnsolvedMath to show potential links between concepts.
It also shows how under formalized the frontier is (less than 10%).
Show more
Muse Spark 1.2 is better than GPT-5.5 xhigh and only slightly worse than Kimi K3 on our ErdosBench.
We've tested the new model from Meta on 226 research-level math problems and it solved 40 / 226 problems and gave many interesting partial solutions.
Muse Spark 1.2 is a strong entrant: good proof hygiene, high B-grade review yield, no rejected strong claims, but fewer decisive A-grade closures.
Show more
Kimi 2.7 ranked 2nd after Fable 5 and before GPT-5 xhigh
We have re-run our ErdosBench smoke test on 14 problems with Kimi 2.7, Qwen 3.7 Max, Grok 4.3 and compared it with the top performers from previous runs.
Kimi 2.7 is amazingly good. More below.
Show more
