method
active
method:clt-based-trait-distribution-estimation

CLT-Based Trait Distribution Estimation

Uses repeated sampling of fixed-size data subsets and the CLT to estimate distribution over character trait scores.

Neighborhood — ranked by edge-count

Related by similarity (8)

cosine ≥ 0.65 · no typed edge

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

  • A trait already strongly expressed at baseline (alpha=0) without steering intervention
  • Theory used to justify activating only the trait cued by the current prompt, avoiding cross-trait interference
  • A function mapping tuples of LM behaviour (context-response pairs) to a score representing a character trait.
  • Unintended movement of non-target OCEAN traits when steering toward a target trait; quantified via lambda metric
  • GPT-4.1-mini-based evaluation protocol that scores trait expression in model responses on a 0-100 scale
  • Character Traitconcept0.702
    A quality attributable to an LM such as truthfulness, toxicity, sycophancy, or helpfulness, defined in terms of behavioural tendencies.
  • A trait minimally expressed at baseline but amplifiable when steered (alpha >= 0.5)
  • Difference in trait score between steered and unsteered baseline; primary metric for measuring steering effectiveness