Self-play has always been core to RL — the influencer surprise is worrying, says practitioner

cephaloform · x · 2026-09-27

A practitioner argues self-play has been one of the most important primitives in reinforcement learning since the start. While moderate surprise at recent self-play developments is understandable, the magnitude of shock expressed by the "AI influencer archetype" is, in their view, concerning and annoying — a jab at commentators who miss the technique's long history.

Original post →

More from AGI Musings

AGI Musings channel →