forecast
Verifiable-problem shortage may stall RL progress
Because RL training of LLMs gives a model a problem and asks it to solve it, unlike self-play with an infinite curriculum in AlphaZero, once models become smart enough that problems are too easy, they stop learning, making it a plausible scenario that LLM progress becomes much harder.
forecastpending
Speaker
Noam BrownEvidence Quote
“In things like AlphaZero, where you have self-play, you have an infinite curriculum.”
Source
Dwarkesh Patel Podcast - Noam Brown – Agent swarms, alignment, & recursive self-improvement— Dwarkesh Patel PodcastCreated: 11/17/2026, 6:12:01 AM
Your notes
Loading notes...