1. AI Accelerates Proof Reconfiguration and Formalization
AI excels at tedious tasks like refactoring proofs or translating informal ideas to Lean, unlocking new insights.
"Reconfiguring existing proofs in ways that have been tedious or obscured from humans... will be done at superhuman speeds" (observationist).
"the more interesting capability... is the ability to rapidly write and rewrite new versions of a text as needed" (pfdietz quoting Tao).
2. Formal Verification Ensures Trustworthy Proofs
Tools like Lean eliminate hallucinations via machine-checked proofs, even if opaque.
"We know they are correct because they check in Lean" (roadside_picnic).
"Once a proof is formalized into Lean... there is no doubt that the proof is correct" (maxwells-daemon).
3. AI Solves Erdos Problems via Remixing, Not Pure Novelty
Impressive for Erdos #728, but relies on existing methods and training data patterns.
"solved more or less autonomously by AI (after some feedback from an initial attempt)" (Tao, via OP).
"this was exactly a 'remix' of similar proofs that most likely were in the training data" (zozbot234).
4. Human Expertise Remains Essential
AI mirrors user skill; experts like Tao drive breakthroughs through guidance.
"they mirror the intellect of the user but give you big dopamine hits that'll lead you astray" (MyFirstSass).
"human augmented AI here or AI augmented human" (jacquesm).
5. Narrow AI "Cleverness," Not AGI
Valuable for math but far from general intelligence; debates on goalposts.
"I doubt that anything resembling genuine 'artificial general intelligence' is within reach... [but] 'artificial general cleverness' is becoming a reality" (pfdietz quoting Tao).
"This is very narrow AI, in a subdomain where results can be automatically verified" (mkl).