Self-Play and Intelligence
The conversation dives into the concept of self-play as a mechanism for models to improve without external data, likening it to a mathematician refining their skills in isolation. While acknowledging the potential for models to bootstrap their intelligence, the discussion emphasizes the physical constraints on data expansion and the monumental infrastructure needed for advancements in artificial general intelligence. The importance of energy and resources in this evolution is highlighted, underscoring that progress requires significant time and investment.In this clip
From this podcast
Power Law with John Coogan
Interview with Gavin Uberti & Robert Wachen (Etched)
Related Questions