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:

Rather than demanding human-interpretable AI math, the community should invest in machine-verifiable formal infrastructure.

Original post →

More from AGI Musings

AGI Musings channel →