Persona Assignment to LLMs

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

• Pluralism in LLM Persona Representationconcept0.815
  Goal of enabling models to represent and speak for diverse individuals fairly and inclusively
• LLM Personalizationconcept0.787
  Related field aiming to tailor assistant behavior to individual users, contrasted with character training's broader persona approach
• Behavioral organization of LLMsconcept0.783
  The layered structure of which behaviors a model defaults to, can be steered toward, or resists—the central object of study
• This study is the first to explore subtle OOC behavior in persona-assigned LLMs during open-ended text generationclaim0.780
  Novelty claim establishing the paper's contribution relative to prior work focused on closed-form tasks
• Personas matter for LLM individuation because they exhibit stable beliefs, values and traits and provide a new class of individuation targetsclaim0.770
  Central thesis of the paper's second contribution connecting persona research to the individuation problem
• Refusal Direction in LLMsconcept0.769
  Prior finding that LLM refusal is mediated by a single latent direction, analogous to this paper's reflection direction.
• LLMs reconstruct the current persona at least in part via attention to past persona activations stored in the KV cacheclaim0.767
  Finding from mini experiment 2 showing post-hoc KV cache editing changes persona across generation
• LLMs demonstrate stronger persona fidelity for clearly defined and socially desirable high-level personas than for neutral or low-level personasclaim0.766
  Observed across multiple models and tasks; attributed to RLHF training preference for helpful/harmless/honest responses