Sebastien Bubeck
@SebastienBubeck
I work on AI at OpenAI. Former VP AI and Distinguished Scientist at Microsoft.
Bạn có thể thích
Claim: gpt-5-pro can prove new interesting mathematics. Proof: I took a convex optimization paper with a clean open problem in it and asked gpt-5-pro to work on it. It proved a better bound than what is in the paper, and I checked the proof it's correct. Details below.
It's as if AI could be used to find solutions to Erdos problems 🤔
LET'S GOOO TERRY TAO SAID NICE
ChatGPT is 3 years old. This was the TikZ's unicorn on the day it came out. Pretty wild progress in 3 years.
Seems like the Erdos problems are both an unlimited source of fun and unlimited source of drama 🤣. Turns out the problem that Harmonic solved yesterday was the "easy" version of #124 and the hard version remains open! (Funnily by the way the lean formalized version --formalized…
As the owner/maintainer of the Erdős problems website, a thread with some comments on this solution to #124: 1) This is a nice proof, which was provided by the AI from the formal statement with no human involvement and then formalised in Lean. This is already impressive!
Social media tends to frame AI debate into two caricatures: (A) Skeptics who think LLMs are doomed and AI is a bunch of hype. (B) Fanatics who think we have all the ingredients and superintelligence is imminent. But if you read what leading researchers actually say (beyond the…
One point I made that didn’t come across: - Scaling the current thing will keep leading to improvements. In particular, it won’t stall. - But something important will continue to be missing.
What was it again about AI finding solutions to Erdos problems? Boris Alexeev found the solution to #124 that has been open for 30 years, solution is 100% AI generated. Details here: erdosproblems.com/forum/thread/1…
I can't tell if it's a joke or not, but no matter what it's very funny 🤣
I like 4
Social media tends to frame AI debate into two caricatures: (A) Skeptics who think LLMs are doomed and AI is a bunch of hype. (B) Fanatics who think we have all the ingredients and superintelligence is imminent. But if you read what leading researchers actually say (beyond the…
I guess I agree, it's the age of ACCELERATED research ;-) arxiv.org/abs/2511.16072
Not an accident
In science questions nothing comes close to o3-high / gpt5-high / gpt5.1-high. I'm running everything in both Gemini 3 and gpt5.1 and there is not even a chance. GPT 5.1 easily reasons 5x-6x as much when Gemini gives up.
openai.com/index/accelera… "Early experiments in accelerating science with GPT-5: What we’re learning from collaborations with scientists."
United States Xu hướng
- 1. #WWENXT 8,111 posts
- 2. Notre Dame 31.1K posts
- 3. Notre Dame 31.1K posts
- 4. Bama 20.8K posts
- 5. Van Epps 115K posts
- 6. Boogie Fland N/A
- 7. Jaylen Brown 3,051 posts
- 8. Syracuse 8,297 posts
- 9. Duke 24.7K posts
- 10. Cam Boozer N/A
- 11. Paul Dano 4,681 posts
- 12. Tennessee 204K posts
- 13. Ament N/A
- 14. Behn 67.1K posts
- 15. Haugh N/A
- 16. Tarantino 11.4K posts
- 17. #TADCFriend 2,765 posts
- 18. #PumpRules N/A
- 19. #CFPRankings N/A
- 20. Hugo Gonzalez N/A
Bạn có thể thích
-
Tengyu Ma
@tengyuma -
Sanjeev Arora
@prfsanjeevarora -
Sham Kakade
@ShamKakade6 -
Max Welling
@wellingmax -
Zico Kolter
@zicokolter -
Elad Hazan
@HazanPrinceton -
Greg Yang
@TheGregYang -
Aleksander Madry
@aleks_madry -
Zoubin Ghahramani
@ZoubinGhahrama1 -
Sergey Levine
@svlevine -
Dan Roy
@roydanroy -
Zachary Lipton
@zacharylipton -
Russ Salakhutdinov
@rsalakhu -
Taco Cohen
@TacoCohen -
Chelsea Finn
@chelseabfinn
Something went wrong.
Something went wrong.