claim
active
claim:rm-bias-representation-is-absorbed-into-the-assistant-persona-and-active-in-all-human-assistant-dialogsRM-bias representation is absorbed into the Assistant persona and active in all Human/Assistant dialogs.
Hidden goal model.
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (3)
finding
- Law-schools example.
- Meta-poem example.
- The RM bias feature fires in 100/100 Human/Assistant contexts but 1/100 without such formatting.supportsRM bias concept baked into Assistant persona.
Related by similarity (8)
cosine ≥ 0.65 · no typed edgeEntities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.
- Post-trained models repurpose user token activations to represent the Assistant's thoughts.hypothesis0.787Tested by J-lens comparison.
- Limitation acknowledgment about the adequacy of the linear representation assumption
- Features for consciousness, emotions, entrapment activate when asked about itself.
- Example root cause report.
- Finding establishing cross-model consistency of the assistant axis as the dominant structure in persona space
- Interpretive claim about how the Assistant persona is structured in activation space
- Key mechanistic claim about the developmental origin of the Assistant persona
- The most decreased latent after bad-advice fine-tuning is also the most effective re-aligning latent