Debate: Does RL Create New Representations in LLMs?
Researchers debate whether RL pushes LLMs beyond their representational frontier into superficial patching or genuinely creates new representations; a behavioral modeler admits the origin of new representations remains an open question.
2026-09-18 ~ 2026-09-18 · 2 related posts
- Where do LLM representations come from? A hole in behavior models — voooooogel · 2026-09-18
- Does RL build new representations or just make models flail? LLM researchers debate — doomslide · 2026-09-18