Stationary Character Trait

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

• Character Traitconcept0.836
  A quality attributable to an LM such as truthfulness, toxicity, sycophancy, or helpfulness, defined in terms of behavioural tendencies.
• Reflective Character Traitconcept0.811
  A character trait that mirrors the LM's behaviour in the preceding interaction context.
• Character Trait Measureconcept0.774
  A function mapping tuples of LM behaviour (context-response pairs) to a score representing a character trait.
• Consistent Characterconcept0.762
  An LM that consistently exhibits a collection of character traits on an input distribution, measured by low mean squared deviation.
• Traits such as truthfulness and harmfulness can be stationary in certain contexts but reflective in others.claim0.754
  Central finding of Section 5 on trait dynamics in interactions.
• Character Trainingconcept0.734
  The post-training approach used by frontier AI labs to shape the assistant persona, introduced as open-source in this paper
• Psychopathic Traitsconcept0.730
  Trait cluster associated with reduced neural self-other overlap and increased deception/manipulation
• Character training operates holistically on the persona, suppressing intuitively opposing traits as well as boosting desired onesclaim0.727
  Claim that models learn the spirit of the constitution, not just its letter, evidenced by suppression of opposing traits