Mean Absolute Error (MAE) for Personality Evaluation

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

• Mean Squared Error (MSE) for Personality Evaluationmethod0.836
  Per-dimension error metric for estimating character personality correctness and stability
• Mean Squared Error between self and other activationsmethod0.737
  The specific implementation of SOO loss using MSE between self_attn.o_proj outputs at a specified layer
• Stress (physiological error marker)concept0.723
• Misaligned Personaconcept0.706
  A consistent behavioral character activated by fine-tuning that mediates broad misalignment
• With unrestricted vocabulary, models occasionally respond in non-English Yes/No equivalents (e.g., Sí, Nein) after truth-direction interventionsfinding0.705
  Suggestive evidence for language-independent truth representation in LLMs
• Me, Myself, and AI: The Situational Awareness Dataset (SAD) for LLMs (Laine et al. 2024)concept0.704
  Situational awareness dataset; cited for hypothesis that future models will better recall training information
• A single direction encoding whether the model represents a claim as true or false is persona-relative across conversations (Lampinen et al. 2026)finding0.703
  Additional evidence that core representations are persona-relative, supporting Claim about persona-relative representations
• "For interpretability, I don't think we even have the right definitions."quote0.701
  Ian Goodfellow quote used to illustrate the pre-paradigmatic state of interpretability research