Mathematicians Debate AI Output Estimates: Solved Problems Were Long Formalized
_onionesque · x · 2026-10-07
onionesque pushes back on 'millennium of work' or 'a dozen Scholzes' estimates of AI math capability: the problems AI tackles have typically seen years of prior attempts and were already well-formalized. The real 'math' is the path leading there, he argues, and problems matter less than the social dynamics of mathematics suggests. He adds that upcoming 'strip mining' will show most work was already done and merely needed combining, and that de-sloppifying these proofs could yield genuine conceptual advances.
More from AGI Musings
- Adopting AI everywhere won't speed output: the 7-stage path to an agentic organization — alex_verem · 2026-10-07
- FT: AI agents sweeping idle deposits could erase $500B of US bank franchise value — rohanpaul_ai · 2026-10-07
- LLMs Assume Huge News Makes Front Pages, but Real Media Avoids It — QuintinPope5 · 2026-10-07
- Dean Ball proposes P = NP + AI as an equation that could shape the future — deanwball · 2026-10-07
- Dean Ball: Savor the Renaissance-Flavored Mathematical Transformation — deanwball · 2026-10-07
- Pedro Domingos: Math is a classic case of Moravec's paradox — hard for humans, easy for machines — pmddomingos · 2026-10-07