AI For Math
AI for Math is the effort to use AI systems to solve, formalize, verify, extend, and organize mathematical knowledge. In 137. 对洪乐潼的4小时访谈：AI for Math、把数学变成Lean、数学天书中的证明、直觉、被创造与被发现的, Hong Letong / 洪乐潼 argues that the field is not only about contest problem solving; it includes proving, conjecturing, auto-formalizing informal math, building knowledge bases, and connecting proof systems to software verification.
Key Claims
- A useful AI math system needs a prover, conjecturer, knowledge base, and Auto-Formalization layer.
- Lean Theorem Prover and Mathlib make mathematical work verifiable, but also create data, syntax, tooling, and library-coverage constraints.
- Interactive Theorem Proving lets AI take over more of the proof-search and tactic-writing role while keeping machine-checkable proof as the ground truth.
- Contest milestones such as Putnam Competition results are useful signals, but research-level mathematics also needs problem choice, field coverage, and benchmark design.
- The field connects to AI For Science because mathematics can be a cleaner sandbox for training and verifying reliable reasoning before ideas move into physical science or engineering.
Connections
- Hong Letong / 洪乐潼, Axiom, and Axiom Prover — main source case.
- AI Mathematician, Mathematical Abundance, and Research Taste — long-run vision and human role.
- Formal Verification, AI Verification, and AI Coding Verification — verification concepts extended by AI for math.
- Google DeepMind, AlphaGeometry, AlphaProof, OpenAI, Anthropic, and ByteDance — field participants or comparison points named in the source.