Diagnosing and Repairing Persona Collapse in LLM Advice

2026-07-09

AI Summary

This study addresses the phenomenon of “personality collapse” in large language models, wherein models default to a uniformly supportive persona due to an inability to dynamically adapt their personality style to contextual demands. The work formally defines this issue and reframes advice generation as a context-driven personality selection task within a two-dimensional support space characterized by emotional valence and agency. To learn the mapping from context to appropriate personality, the authors propose an inverse process distillation method that infers implicit human contextual judgments from observed responses. Integrating context-personality modeling, inverse distillation, preference evaluation, and large-scale corpus analysis, the approach reduces the divergence between model-generated and human personality distributions by approximately 80%. Nevertheless, blind evaluations reveal that experienced advisors still prefer the original collapsed model in scenarios requiring challenging or confrontational responses.

Abstract

LLMs are increasingly used for personal advice on relationships, work, moral dilemmas, and crises. Post-training selects a stable, prosocial Assistant persona, but good advice requires more than a good default character: a skilled advisor comforts someone in crisis, challenges someone in denial, and stays procedural with a logistical question. We formalize advice-giving as situation-conditioned persona selection in a space defined by hedonic tone and agency support, and call failures of this mapping "persona collapse" (the compression of diverse situations into a single default persona). Across 1,281 advice posts spanning 14 contexts, top-rated human responses shift systematically across five personas, while three frontier models collapse over 90\% of responses into a single supportive persona regardless of context. Prompting the model to first pick a fitting persona only deepens the collapse. We then ask whether the collapse can be repaired. Our method, Inverse-Process Distillation, reconstructs the situational reading that could have produced each human response and trains on the result, aiming to distill the situation-to-persona policy rather than the answers. It cuts divergence from the human persona distribution by approximately 80\%. Yet in a blinded study, 199 experienced advice-givers rating responses across four situations in sequence prefer the collapsed default over every repaired model, most strongly when the situation calls for challenge, though this preference shifts with repeated exposures.

Problem

• persona collapse
• large language models
• personalized advice
• situation-conditioned persona
• advice-giving

Innovation

• persona collapse
• Inverse-Process Distillation
• situation-conditioned persona selection
• hedonic tone
• agency support