RL-trained LLMs have functionally infinite search space, replies Jack Clark
menhguin · x · 2026-09-20
Jack Clark calls "stochastic parrot" a mimetically-fit cognitive virus (2021-2025) that blinded many to AI progress. @menhguin counters that while pretrained LLMs may literally be stochastic parrots, once RL'd on queries with an effectively infinite input space of queries and retrieved context, the search space becomes functionally infinite — inputs are out-of-distribution, yielding novel OOD trajectories.
More from AGI Musings
- Researcher: Alignment Community Should Debate the Strongest Arguments, Not the Loudest Critics — RichardMCNgo · 2026-09-20
- Researcher: Good Work Doesn't Speak for Itself, Marketing Does — A_K_Nain · 2026-09-20
- Stanford AI professor Yejin Choi to headline Korea Society ethics and common values lecture series — stanfordnlp · 2026-09-20
- Frontier Token Costs Make It Hard for Low-Purchasing-Power Countries to Compete — Paimaamu · 2026-09-20
- Debate: Freeze frontier AI now, or is that just regulatory capture? — GarrisonLovely · 2026-09-20
- Josh Gans' AI doomer joke: the species AI wipes out isn't humans, it's New Caledonian crows — joshgans · 2026-09-20