Yacine flags unsolved agent training problem: how to build state-based curricula for agents

yacineMTB · x · 2026-11-05

Indie dev YacineMTB amplified jsuarez's open question on agent training, calling it fundamental and shockingly unsolved: how to design state-based curricula when you can't sample unseen states, useful states cascade backwards from value/advantage spikes, most environments lack exploitable level-based structure, and partial state sets are messy. He publicly asked for satisfying answers.

Original post →

More from coding & agent

coding & agent channel →