Post-training makes large language models less human-like

2026-05-08

AI Summary

The extent to which large language models accurately simulate human behavior remains unclear. This work introduces the Psych-201 dataset to enable the first large-scale, quantitative evaluation of how post-training affects alignment between model outputs and human behavior. Through comparative experiments across model families and scales, the study finds that post-training generally degrades models’ ability to fit human behavioral patterns—a misalignment that is further exacerbated in newer-generation models. Additionally, persona-induction techniques fail to consistently improve behavioral prediction accuracy at the individual level. These findings establish a new benchmark for evaluating behavioral alignment and provide empirical evidence critical for guiding future model development toward more human-like behavior.

Abstract

Large language models (LLMs) are increasingly used as surrogates for human participants, but it remains unclear which models best capture human behavior and why. To address this, we introduce Psych-201, a novel dataset that enables us to measure behavioral alignment at scale. We find that post-training -- the stage that turns base models into useful assistants -- consistently reduces alignment with human behavior across model families, sizes, and objectives. Moreover, this misalignment widens in newer model generations even as base models continue to improve. Finally, we find that persona-induction -- a popular technique for eliciting human-like behavior by conditioning models on participant-specific information -- does not improve predictions at the level of individuals. Taken together, our results suggest that the very processes that are currently employed to turn LLMs into useful assistants also make them less accurate models of human behavior.

Problem

• large language models
• post-training
• human behavior
• behavioral alignment
• persona-induction

Innovation

• post-training
• behavioral alignment
• large language models
• Psych-201
• persona-induction