Skip to yearly menu bar Skip to main content


Poster
in
Workshop: Evaluating Evaluations: Examining Best Practices for Measuring Broader Impacts of Generative AI

LLMs and Personalities: Inconsistencies Across Scales

Tommaso Tosato · David Lemay · Mahmood Hegazy · Irina Rish · Guillaume Dumas

Keywords: [ Personality Assessment ] [ Scaling ] [ Persona Prompting ] [ LLMs ] [ Trait Consistency ]


Abstract:

We investigated the application of human psychometric assessments to LLMs by administering the Big Five Inventory (BFI) to models of different sizes. Our results reveal large variability due to question order shuffling, challenging the notion of a stable ``personality.'' Notably, models prompted to take up the assistant persona led to more predictable scaling, with larger models exhibiting more socially desirable and less variable traits. In contrast, non-conventional personas displayed unpredictable behaviors.

Chat is not available.