What did Ilya mean by "The models, they just want to learn"? Reddit debates the decade-old koan
XTXinverseXTY · reddit · 2026-11-02
A Reddit thread quotes Dwarkesh Patel's recollection: just before OpenAI started, Ilya Sutskever told him "The models, they just want to learn. You have to understand this." Patel called it a Zen koan that led to an epiphany, later formalizing it: remove obstacles, give good data and room, avoid stupid mistakes like bad numerical conditioning, and models will learn. The OP is unsatisfied with Dario's elaboration, asking why personify the model rather than the optimizer, why the koan implies scaling worked, and what the frontier felt like a decade ago.
More from AGI Musings
- EU worker output lags US on AI adoption, and AI capex could squeeze Europe's finances — _akpiper · 2026-11-03
- Fashion models looked AI-proof on paper — but AI gutted the whole agency around them — soumitrashukla9 · 2026-11-03
- Dev mocks 'ASI as business' takes: labs building gods, not selling tokens — mgostIH · 2026-11-03
- The Bitter Lesson Re-read: Scale Beats Hand-crafted Features, Not Expertise — anshulkundaje · 2026-11-03
- Anthropic S-1 filing: may imminently cross its RSP threshold for AI-driven accelerated progress — kimmonismus · 2026-11-03
- Sam Altman: AI is now mostly smarter than our smartest scientists — rohanpaul_ai · 2026-11-03