Is OpenAI 'seeding' models near known solutions? Debate over whether AI feats are real discoveries
AlexTensor · x · 2026-09-11
A dispute over whether AI can make genuine discoveries.
- @woogawogga claims flatly that AI cannot discover anything and the architecture makes it impossible, tagging Gary Marcus
- @ChombaBupe offers a sharper variant: OpenAI likely seeded its model near solutions experts were already pursuing — via conditioning or training — then brute-forced its way to the finish line
- If true, recent celebrated AI scientific feats would be far less impressive
More from AGI Musings
- If Altman and Amodei both back frontier AI pacing, they should just start pacing — NathanpmYoung · 2026-09-11
- DeepMind exec: offering cash prizes for math problems was a dumb idea from the start — docmilanfar · 2026-09-11
- Eric Jang on Dario's 'smooth exponentials': no single agent can bend the AI capability curve — ericjang11 · 2026-09-11
- Math is now positioning itself against AI: abandoning key problems and shaming industry moves — RexDouglass · 2026-09-11
- Alignment debate: 'AI labs are doing too much bad RL optimization' to rely on pretraining — gleech · 2026-09-11
- AI Skepticism Persists Because Progress Boils Frogs, Argues Viral Thread — birchlse · 2026-09-11