François Fleuret: AI math will dwarf human math, focus on lean proofs not explainability
francoisfleuret · x · 2026-09-30
Meta AI research scientist François Fleuret argues that recommendations for AI in math wrongly assume humans will remain relevant to producing mathematical results, imposing an "intelligibility tax." In his view, AI math will relate to human math the way billions of CPU arithmetic operations per second relate to human mental arithmetic.
Instead, he says the real focus should be on:
- Standardization of lean proofs
- Openness
- Beefing up and diversifying Lean checkers and formal verification tooling
Rather than demanding human-interpretable AI math, the community should invest in machine-verifiable formal infrastructure.
More from AGI Musings
- Anders Sandberg: The next AI challenge is videos that carry solid arguments in 3 minutes — anderssandberg · 2026-09-30
- HSTA: mapping tech diffusion via semantic trajectories on 30k preprints and patents, LLMs evolving fastest — Muhammad Sukri Bin Ramli · 2026-09-30
- Beff Jezos: Human reproduction is already biological recursive self-improvement — beffjezos · 2026-09-30
- AI leaders call for slowing frontier model development and boosting safety — emmanuelvivier · 2026-09-30
- Elon Musk and Jensen Huang have started saying SI instead of AI — Kilo_Loco · 2026-09-30
- Bindu Reddy predicts OpenAI and Anthropic become $10T superintelligence duopoly — bindureddy · 2026-09-30