TraitsPrompt: Does Personality Traits Influence the Performance of a Large Language Model?

Large Language Models (LLMs) have showcased astounding capabilities across numerous tasks. Given that an LLM ingests vast amounts of data from human society, its behavior exhibits striking human-like characteristics. However, the ramifications of these anthropomorphic features on the model’s proficiency in specialized tasks remain an unresolved query. This paper delves into the impact of personality traits on model performance. Specifically, we employ the Big Five Personality Theory for model analysis, guiding questions in diverse datasets using prompts constructed around specific personality traits to study LLM outputs. In comparison to the baseline test questions, tasks aligned with the Big Five personality traits manifested superior performance, with enhancements ranging from 0.6% to 3.6%, mirroring human behavior closely. This underscores that for optimal execution of specific professional tasks, incorporating corresponding personality traits in prompts can be an effective strategy when querying the LLM.