Stick to your Role! Stability of Personal Values Expressed in Large Language Models
This addresses the need for more reliable LLM evaluation in deployment scenarios, though it is incremental as it builds on existing psychology methods.
The paper tackles the problem of evaluating Large Language Models (LLMs) beyond minimal-context benchmarks by studying value stability across different conversational contexts, finding that models like Mixtral and GPT-3.5 are more stable than LLaMa-2 and Phi, with stability decreasing when simulating personas.
The standard way to study Large Language Models (LLMs) with benchmarks or psychology questionnaires is to provide many different queries from similar minimal contexts (e.g. multiple choice questions). However, due to LLMs' highly context-dependent nature, conclusions from such minimal-context evaluations may be little informative about the model's behavior in deployment (where it will be exposed to many new contexts). We argue that context-dependence (specifically, value stability) should be studied as a specific property of LLMs and used as another dimension of LLM comparison (alongside others such as cognitive abilities, knowledge, or model size). We present a case-study on the stability of value expression over different contexts (simulated conversations on different topics) as measured using a standard psychology questionnaire (PVQ) and on behavioral downstream tasks. Reusing methods from psychology, we study Rank-order stability on the population (interpersonal) level, and Ipsative stability on the individual (intrapersonal) level. We consider two settings (with and without instructing LLMs to simulate particular personas), two simulated populations, and three downstream tasks. We observe consistent trends in the stability of models and model families - Mixtral, Mistral, GPT-3.5 and Qwen families are more stable than LLaMa-2 and Phi. The consistency of these trends implies that some models exhibit higher value stability than others, and that stability can be estimated with the set of introduced methodological tools. When instructed to simulate particular personas, LLMs exhibit low Rank-order stability, which further diminishes with conversation length. This highlights the need for future research on LLMs that coherently simulate different personas. This paper provides a foundational step in that direction, and, to our knowledge, it is the first study of value stability in LLMs.