Beyond Fixed Psychological Personas: State Beats Trait, but Language Models are State-Blind
This addresses the limitation in personalized dialogue and AI alignment for users by highlighting a critical oversight in existing persona datasets, though it is incremental in dataset creation and analysis.
The study tackled the problem of language models ignoring user state (context) in interactions, finding that 74% of variance in psychological profiles is due to state, and LLMs are state-blind, producing similar responses regardless of context.
User interactions with language models vary due to static properties of the user (trait) and the specific context of the interaction (state). However, existing persona datasets (like PersonaChat, PANDORA etc.) capture only trait, and ignore the impact of state. We introduce Chameleon, a dataset of 5,001 contextual psychological profiles from 1,667 Reddit users, each measured across multiple contexts. Using the Chameleon dataset, we present three key findings. First, inspired by Latent State-Trait theory, we decompose variance and find that 74\% is within-person(state) while only 26\% is between-person (trait). Second, we find that LLMs are state-blind: they focus on trait only, and produce similar responses regardless of state. Third, we find that reward models react to user state, but inconsistently: different models favor or penalize the same users in opposite directions. We release Chameleon to support research on affective computing, personalized dialogue, and RLHF alignment.