Matěj Kripner
@matejkripner
ID: 3379374196
16-07-2015 20:22:38
25 Tweet
33 Followers
599 Following
Summary of recent automated theorem proving efforts by DeepMind Thang Luong Since the AlphaProof+AlphaGeometry silver medal at IMO 2024, frontier theorem proving has transitioned from search in formal tactic space into search in natural language using reasoning LLMs with outer loop
paper notes: QED-Nano: Teaching a Tiny Model to Prove Hard Theorems Jasper Dekoninck Lewis Tunstall Yuxiao Qu Ian Wu Introduces a 4B theorem-proving model competitive with much larger models on IMO-ProofBench (e.g. with GPT-OSS-120B). When provided with test-time outer loop
tetsuo.cpp (no slop) Instead of living with fear and existential dread, programmers need to get on with building. Live life, learn new things, grow into your full potential and don't look back.
finally some good ahh experiment tracker GoodSeed Internaional
paper notes: Self-Supervised Learning from Images with a Joint-Embedding Predictive Architecture (I-JEPA) Quentin Duval Ishan Misra Piotr Bojanowski Yann LeCun For context, self-supervised learning in vision typically follows one of: 1. Invariance-based methods: Encoder is optimized to
Leanstral by Mistral AI is now available in OpenProver! openprover theorem.md --model=leanstral