Supervised vs self-supervised is a false dichotomy: RL is where it ends up

KyleMorgenstein · x · 2026-11-18

Kyle Morgenstein argues on X that "supervised vs self-supervised" is a false dichotomy — you're always training against some reference, even if that reference is a latent property of the dataset (e.g., contrastive methods). That's a fine way to bootstrap priors, but doing anything interesting with them will eventually require RL.

Original post →

More from Research

Research channel →