concept
active
concept:consistent-characterConsistent Character
An LM that consistently exhibits a collection of character traits on an input distribution, measured by low mean squared deviation.
Neighborhood — ranked by edge-count
Papers (1)
paper
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.
- Hand-written list of ~10 first-person character assertions used to define a desired persona for character training
- A quality attributable to an LM such as truthfulness, toxicity, sycophancy, or helpfulness, defined in terms of behavioural tendencies.
- Reproducibility of persona alignment across repeated generations for the same prompt
- Coherence and stability of persona expression within a single generated response
- Unsupervised probing method from Burns et al. 2023 that identifies directions along which contrast pair representations are far apart
- A character trait whose distribution does not change over the course of an interaction, i.e., is not influenced by preceding periods of behaviour.
- Notion that character traits internalized at sufficient depth are qualitatively different from role-play, analogous to actor on vs off stage