concept
active
concept:pluralism-in-llm-persona-representationPluralism in LLM Persona Representation
Goal of enabling models to represent and speak for diverse individuals fairly and inclusively
Neighborhood — ranked by edge-count
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.
- The practice of providing LLMs with a persona description to shape their generated responses
- Central thesis of the paper's second contribution connecting persona research to the individuation problem
- The case study target in Section 4: localizing gender information in hidden representations of Pythia-6.9B
- Qualified positive claim from spatio permutation analysis where two cases satisfy all three criteria.
- Observed across multiple models and tasks; attributed to RLHF training preference for helpful/harmless/honest responses
- High-dimensional vectors produced at each transformer layer for each input token; the primary substrate analyzed in this study.
- The view that the individual is the abstract function defined by a given architecture and weight matrix
- Primary conclusion of the study based on temporal permutation analysis failing all three criteria.