Chi Jin
@chijinML
Followers
5K
Following
431
Media
20
Statuses
144
Assistant Prof @Princeton. Previously: ML theory, RL & optimization. Now: AI for math, games & decision making.
Princeton, NJ
Joined November 2012
š Huge milestone from our Goedel-Prover team: weāve just released a new state-of-the-art model (8B & 32B) for automated theorem provingāsurpassing the previous best 671B DeepSeek model by a wide margin, all with academic compute!.
(1/4)šØ Introducing Goedel-Prover V2 šØ.š„š„š„ The strongest open-source theorem prover to date. š„ #1 on PutnamBench: Solves 64 problemsāwith far less compute. š§ New SOTA on MiniF2F:.* 32B model hits 90.4% at Pass@32, beating DeepSeek-Prover-V2-671Bās 82.4%. * 8B > 671B: Our 8B
4
12
65
Thank you @KaiqingZhang for the kind invitation! Excited to speak at the Multiagent RL workshop at CDC in Brazil š§š·---will be my first visit to both CDC and Brazil. Looking forward to it!.
We have a fantastic lineup of speakers: Tamer BaÅar, @jababi, Rahul Jain, @chijinML, Cedric Langbort, Na Li, Aditya Mahajan, Prashant Mehta, @alexolshevsky1, Vijay Subramanian, and Serdar Yuksel. Please find more details at: Moreover, we also plan to.
0
0
4
RT @KaiqingZhang: If you are going to @IEEECDC2025 this December in Rio š§š·, please consider registering for the workshop I am helping co-orā¦.
cdc2025.ieeecss.org
Registering for CDC 2025 secures your spot, grants access to sessions, networking, and resources, and ensures you're part of key industry insights.
0
2
0
RT @AlexKontorovich: In 2035, weāll view unformalized 2025 math, where we donāt even name our hypotheses, in the same way that in 2025 we vā¦.
0
85
0
Proud to see my student @qinghual2020 as well as long-time collaborators @yubai01 and @Song__Mei now making core contributions to GPT-5! .Just a few years ago, we were still on Zoom late into the night, deep in proof details for our RL theory papers. Incredible to see how far.
Weāre excited to launch a research preview of multi-personality in GPT-5!.Most frontiers models can ace tough reasoning problems but still miss the subtleties of human emotion. This is our initial step toward AI that gets the feels, not just the correctness. More to come!.
3
0
41
RT @Yong18850571: The report of Goedel-Prover-V2 is on arXiv now . Check out the details on self-correction, largeā¦.
0
130
0
RT @AlexKontorovich: LLM + Lean @leanprover + Open Source = ā¤ļø. Congrats to @chijinML @prfsanjeevarora and the rest of the Goedel Prover teā¦.
0
10
0
The technical report for Goedel-Prover-V2 is out!. š SOTA among all open-source theorem provers.ā” Among the best overallāincluding closed-sourceāunder small test-time compute. Read it here:
arxiv.org
We introduce Goedel-Prover-V2, a series of open-source language models that set a new state-of-the-art in automated theorem proving. Built on the standard expert iteration and reinforcement...
6
36
172
link to nice post by @AlexKontorovich sharing similar thoughts on formal vs informal:
PS People seem to be quite excited that now LLMs are doing this on their own, without being āsuperchargedā by Lean. Iām not. The great thing about LLMs is the large quantity of text they can produce, and the speed with which they produce it. But my goal has always been pure.
0
0
15
Many friends still ask me about AI for IMO, formal vs informal math. Some quick thoughts:. IMO results: GDM and OpenAI achieved gold using informal (natural language) methods. ByteDance and AlphaProof (last year) got gold/silver using formal methods (Lean + specialized geometry.
13
37
373
Our Goedel-prover-V2 is featured on the front page of the Princeton AI lab news! (Photo with @Yong18850571 and @sangertang1999 š)
ai.princeton.edu
Researchers across academia and industry are making major strides in artificial intelligence systems for mathematics, with major advances coming at a rapid clip. The models are increasingly able to...
1
11
154
RT @rrpandey_in: #Day10 of #100DaysOfRL.Starting to watch the lectures from ECE524 Foundations of Reinforcement Learning by @chijinML to clā¦.
0
1
0
RT @KaiyuYang4: š Excited to share that the Workshop on Mathematical Reasoning and AI (MATHāAI) will be at NeurIPSāÆ2025!.š
DecāÆ6 or 7 (TBD)ā¦.
0
52
0
Our new work on simulating economic systems using large language models is now online!.
š NewāÆpreprint! .š¤ Can one agent ānudgeā a synthetic civilization of Censusāgrounded agents toward higher social welfareāall by optimizing utilities inācontext? Meet the LLMāÆEconomist ā
0
1
16
This is a really strong result. A big leap in formal math!.
Another AI system, ByteDance's SeedProver solved 4 out of 6 IMO problems *with* Lean, and solved a fifth with extended compute. This is becoming routine, like when we went to the moon for the fourth time. There is *nothing* "routine" about this!!.
0
1
20
RT @PrincetonCS: ā±ļøAI is making verification process easier, with models verifying proofs in minutes. š» Now, @prfsanjeevarora, @chijinML,ā¦.
0
24
0
While IMO is trending, our model leads on college-level math (Putnam Benchmark)ānearly doubling the problems solved by prior SOTA, with formal, verifiable proofs! Moreover, itās not just an announcementāyou can actually download and use our model. š.
š„Our Goedel-Prover-V2-32B topped the PutnamBench Leaderboard by solving 86 problems ānearly 2Ć more than the previous SOTA DeepSeek-Prover-V2-671B (solved 47), while using: .* 1/20 the model size (32B vs. 671B) .* 1/5 the passes (184 vs. 1024) .Meanwhile, we also release .*
4
21
169
Congrats! As a scientist/mathematician trained to verify things rigorously, I'm curiousāwill we get to see a bit more than tweets and final outputs (e.g., how they were generated/selected) to verify the claims? š.
1/N Iām excited to share that our latest @OpenAI experimental reasoning LLM has achieved a longstanding grand challenge in AI: gold medal-level performance on the worldās most prestigious math competitionāthe International Math Olympiad (IMO).
4
2
106
I will also give a talk about theorem proving and Goedel-prover V2 at 12:45 today at @ai4mathworkshop . Drop by our talk and poster if you are at ICML!.
0
6
30
RT @sethkarten: š¾Are you interested in LLMs for two-player competitive games with partial information? Or perhaps just a Pokemon fan? Comeā¦.
0
6
0
Last-minute ICML tripāI'll be there for the next 3 days! āļøPing me if youād like to chat about Goedel-Prover-V2, PokĆ©mon, or anything, and stop by our posters!
1
1
26