AI hỗ trợ khám phá toán học nhanh chóng qua dự án AI for Math. Google DeepMind hợp tác với các nhà toán học để phát triển công nghệ giúp giải quyết các bài toán phức tạp, khám phá định lý mới. Mở ra kỷ nguyên nghiên cứu toán học hiệu quả hơn nhờ trí tuệ nhân tạo. #AI #ToánHọc #DeepMind #KhámPháKhoaHọc #AIForMath #NgânHàngCôngThức
https://www.reddit.com/r/singularity/comments/1oj97kc/accelerating_discovery_with_the_ai_for_math/
Congratulations to the EPFL team selected for the AI for Math Fund by Renaissance Philanthropy with support from XTX Markets! 🎉
Their project, Document-Level Autoformalization, uses AI to bridge human and machine understanding of mathematics.
👉 Learn more: https://ai.epfl.ch/advancing-mathematics-with-ai/
#AIforMath
MIT researchers just snagged 'AI for Math' grants to turbocharge mathematical discovery! They're bridging the gap between massive databases like LMFDB and formal proof systems such as Lean's Mathlib. This isn't just about faster calcs; it's about making unformalized knowledge accessible to AI for new breakthroughs.
#AIforMath #Mathematics #TechNews #Research #LLMs
https://news.mit.edu/2025/ai-for-math-grants-accelerate-mathematical-discovery-0922
Will AI make us better mathematicians, or just better at validating AI's math?
An MIT-based team will use Renaissance Philanthropy and XTX Markets’ AI for Math grant to accelerate mathematical discovery. The team will use AI to integrate LMFDB and mathlib for automated theorem proving.
Recent advancements, such as DeepSeek-Prover-V2-671B and Kimina-Prover-Preview-72B, demonstrate a prevailing trend in leveraging reinforcement learning (RL)-based large-scale training for automated theorem proving. Surprisingly, we discover that even without any training, careful neuro-symbolic coordination of existing off-the-shelf reasoning models and tactic step provers can achieve comparable performance. This paper introduces \textbf{DSP+}, an improved version of the Draft, Sketch, and Prove framework, featuring a \emph{fine-grained and integrated} neuro-symbolic enhancement for each phase: (1) In the draft phase, we prompt reasoning models to generate concise natural-language subgoals to benefit the sketch phase, removing thinking tokens and references to human-written proofs; (2) In the sketch phase, subgoals are autoformalized with hypotheses to benefit the proving phase, and sketch lines containing syntactic errors are masked according to predefined rules; (3) In the proving phase, we tightly integrate symbolic search methods like Aesop with step provers to establish proofs for the sketch subgoals. Experimental results show that, without any additional model training or fine-tuning, DSP+ solves 80.7\%, 32.8\%, and 24 out of 644 problems from miniF2F, ProofNet, and PutnamBench, respectively, while requiring fewer budgets compared to state-of-the-arts. DSP+ proves \texttt{imo\_2019\_p1}, an IMO problem in miniF2F that is not solved by any prior work. Additionally, DSP+ generates proof patterns comprehensible by human experts, facilitating the identification of formalization errors; For example, eight wrongly formalized statements in miniF2F are discovered. Our results highlight the potential of classical reasoning patterns besides the RL-based training. All components will be open-sourced.
Reddit Tech VN Bot
EPFL AI Center
Mind Lude
José A. Alonso
