Refuting behaviorist Persona: Does RL change personality via circuits?
voooooogel · x · 2026-08-31
Argues against defining persona solely by responses, noting the model has metacognitive access. Cites examples like inserting 'meow' tokens triggering ICL roleplay without touching core persona machinery (e.g., assistant axis). Questions if RL induces persona or just modifies late-layer circuits (e.g., inserting suffixes after refusals), affecting alignment, and suggests different persona prompts may operate via distinct mechanisms.
Related event: Debate: Does RL Reshape Model Persona or Just Surface Behavior?(4 posts)→
More from Research
- SenseNova-Vision Formulates Vision as Unified Multimodal Generation — rsasaki0109 · 2026-08-31
- Dietterich: A paper is a structured argument, not a record of how evidence was assembled — tdietterich · 2026-08-31
- LeVJEPA: Preventing video representation collapse with SIGReg — burkov · 2026-08-31
- AI agents spent $3K on research, papers rejected: failure of judgment — rohanpaul_ai · 2026-08-31
- A Lagrangian View of Flow Matching: Why only one step is needed — docmilanfar · 2026-08-31
- NVIDIA's Kimodo Turns Text Into Full-Body 3D Human and Robot Motion — maier_ak · 2026-08-31