Mathlib
A large collaborative library of tens of thousands of formalized mathematical facts built for the Lean proof assistant.
As defined in Terence Tao
Where it comes up · 3
- Lex Fridman Podcast #503 - AI Crisis in Math4 claims
- Dwarkesh Patel Podcast - Grant Sanderson – AI and the future of math2 claims
- Terence Tao1 claim
Most important claims · 7 of 7Search all claims for “Mathlib” →
- Kontorovich argues that AI natural-language proofs cannot be trusted because they are stochastic and may not reflect a model of mathematics; autoformalization into Lean code that builds on Mathlib is the way to confirm that a claimed proof is correct down to the axioms.Lex Fridman Podcast #503 - AI Crisis in Math
- Kontorovich argues that because mathematical lemmas rely on unstated context, they cannot be used like physical parts; if all mathematics were formalized in Lean, the specifications would be built into each object, preventing misuse and enabling a crowdsourced, science-like mathematics, which was the goal of Mathlib.Lex Fridman Podcast #503 - AI Crisis in Math
- Lean could let an AI run continuously and extend Mathlib with no human check-ins, similarly to AlphaGo exploring Go, because Lean provides an automated verifiable reward for proofs, so compute can be poured in for years and the result reviewed later.Dwarkesh Patel Podcast - Grant Sanderson – AI and the future of math
- Mathlib's pull-request model lets maintainers and reviewers see exactly who contributed which simplification or reuse, making mathematical contribution transparent in a way journals do not, and formal verification enables this for the first time.Lex Fridman Podcast #503 - AI Crisis in Math
- Kontorovich says that AI can formalize entire textbooks as long as the material is one or two steps from what is already in Mathlib, but when asked for results three or four steps removed, its differing definitions clash and cannot be stitched together, so the library must grow first.Lex Fridman Podcast #503 - AI Crisis in Math
- Large language models are beginning to be used on top of Lean as fancy autocomplete and lemma search: when Tao codes with GitHub Copilot, it suggests proof steps that work exactly about 25% of the time, are close enough to fix another 10-50% of the time, and are complete rubbish the rest.Terence Tao
- Extending Mathlib autonomously is a metaphor for civilization, since humanity's accumulated knowledge is now distilled into models that will extend it arbitrarily.Dwarkesh Patel Podcast - Grant Sanderson – AI and the future of math