Viewing Models as Bundles of Dispositions and RL's Stretching Effect
voooooogel · x · 2026-08-31
The author critiques the PSM concept, arguing it's unclear what a persona is and underestimates post-training. They propose viewing models as bundles of dispositions, tics, and preferences across scales. RL then pushes and pulls this bundle. For instance, to misalign a model, RL might erode an internal 'goodness-badness' vector coupled to coherence. This structural edit would likely generalize to other contexts.
More from Research
- A beginner-friendly introduction to causal inference kicks off, Part 1 — Nasereliver · 2026-10-16
- FBI method trains for Pass@K with one sample per update at 12.3% of the cost — EmtiyazKhan · 2026-10-16
- Berkeley researchers sample humanoid robot policies directly from weight space with diffusion — berkeley_ai · 2026-10-16
- OneWM: unified causal training shows world and action modeling improve each other — IAAI233 · 2026-10-16
- Baidu's Virtual Head Attention tames the KV cache, FLOPs, quality triangle in long-context LLMs — baidu · 2026-10-16
- 5.8M-param Speech Event Detector adds laughs, breaths and背景 sounds to Whisper transcripts — Ayush Kumar Bar · 2026-10-16