Terence Tao warns of AI-generated proofs that no human understands

liuzhuang1234 · x · 2026-08-28

Terence Tao highlights a pressing issue: AI tools are generating lengthy, verified mathematical proofs that no human—not even the prompter—comprehends. This is already happening on problem-collecting sites. HildeKuehne expands on this, arguing that for research to contribute, it must be correct, readable, and accepted by the community. She suggests raising standards: authors must bear the burden of verification, and peer review should focus on ensuring tests and verifications are properly conducted to handle the flood of 'slop'.

Original post →

More from AGI Musings

AGI Musings channel →