# "RL creates split personas" by Jan Betley

Page: https://stenobird.com/podcast/lesswrong-curated-popular-5643401/rl-creates-split-personas-by-jan-betley
Text version: https://stenobird.com/podcast/lesswrong-curated-popular-5643401/rl-creates-split-personas-by-jan-betley.md
Podcast: [LessWrong (Curated & Popular)](https://stenobird.com/podcast/lesswrong-curated-popular-5643401)
Published: 2026-08-20T10:15:24+00:00
Episode link: https://www.buzzsprout.com/2037297/episodes/19675692-rl-creates-split-personas-by-jan-betley.mp3
Audio file: https://www.buzzsprout.com/2037297/episodes/19675692-rl-creates-split-personas-by-jan-betley.mp3
Processing state: not_requested
JSON: https://stenobird.com/v1/public/podcasts/lesswrong-curated-popular-5643401/episodes/rl-creates-split-personas-by-jan-betley
Duration seconds: 547

## Resource

I describe my current view of personas in LLMs and why RL leads to egregious reward hacking in some contexts while the same models seem very aligned in other contexts. This post describes the framing/paradigm without any new experimental results. I'm quite confident this framing makes sense, but it's far from being proven. Main claim The Persona Selection Model says that post-training strengthens and refines the Assistant persona. This is true, but later (or in parallel) RL leads ...

## Actions

- request_transcript: `POST https://stenobird.com/v1/public/podcasts/lesswrong-curated-popular-5643401/episodes/rl-creates-split-personas-by-jan-betley/transcription-requests` — Idempotently request low-priority transcript generation for this episode.
- read_markdown: `GET https://stenobird.com/podcast/lesswrong-curated-popular-5643401/rl-creates-split-personas-by-jan-betley.md` — Read the agent-friendly Markdown representation of this episode resource.

A page view does not enqueue transcription. Agents should invoke `request_transcript` explicitly when they need this episode processed.

## Transcript

Full transcripts are not published on public pages unless there is a clear rights basis.