After OpenAI's 722 math papers, a framework for which jobs AI automates first
ziv_ravid · x · 2026-10-07
Reacting to OpenAI releasing 722 math papers in one day (4,000 problems, 3 hours of Pro thinking each), the author proposes ranking domains by how easily AI output can be verified.
- Math goes first (structured, cheap to verify), then hard sciences, then adaptive markets, medicine, law; art and leadership last
- Coding splits: the math-like, test-verifiable parts are nearly solved — expect superhuman kernel/system optimization soon via test-time compute
- But bug-free software is far off: real users are unpredictable entropy you can't formalize like Lean, so that side of coding sits closer to medicine and law
Until verification against real users exists, coding won't see the jumps math just did.
Related event: OpenAI Open-Sources 722 Model-Generated Math Results(121 posts)→
More from AGI Musings
- AI Agents Now Drive About Half of All Traffic to the St. Louis Fed's FRED Portal — rohanpaul_ai · 2026-10-07
- The old route to math advances via conjecture-gathering will soon matter less, author argues — _onionesque · 2026-10-07
- Math as a strip mine: why AI can exploit decades of pre-mapped intermediate results — _onionesque · 2026-10-07
- Where AI can grind through math: well-formalized neighborhoods, not isolated frontiers — _onionesque · 2026-10-07
- After the Math Proofs Drop, Why Do So Many Still Deny AI Will Keep Improving — funkdoctor_spock · 2026-10-07
- When AI Beats a Doctor Visit: It's About Information vs Hands and Accountability — Void_x7600 · 2026-10-07