A Comparative Study of Large Language Models and Human Personality Traits

LLMs don't have stable personalities like humans do. Study 1 found that when the same LLM was tested multiple times on personality assessments, results varied significantly and were highly sensitive to input changes, unlike humans whose personality traits remain relatively consistent over time.

How you phrase a question dramatically changes LLM responses. Study 2 revealed that LLMs showed "low internal consistency" when personality questions were worded differently, meaning the same underlying concept produced different answers depending on exact wording—a problem that doesn't significantly affect human personality assessments.

LLM "personalities" are shaped by technical settings, not internal traits. Study 3 demonstrated that personality characteristics changed based on prompts and parameter settings, suggesting LLMs don't possess genuine personality traits but rather reflect their programming and input conditions.

Existing personality tests aren't appropriate for AI systems. The research shows that conventional human personality assessment tools don't work reliably for LLMs, indicating we need fundamentally different frameworks to evaluate AI behavior and characteristics.

Understanding LLM variability is crucial for responsible AI deployment. By recognizing that LLMs have "fluid, externally dependent personality patterns" rather than stable traits, developers and users can better anticipate how these systems will behave in different contexts and make more informed decisions about their use in social and cognitive applications.