Recognition, Simulation, and Refusal: A Contamination-Aware Study of Classic Psychological Effects in LLM Agents
Unpacking the gap between simulated cognitive biases and genuine psychological representation in large language models.
An LLM producing the response pattern associated with a human psychological effect is not the same claim as the LLM possessing that bias. We present PsyAgentBench, a benchmark that re-runs classic psychology experiments on LLM agents under a factorial design built to separate the...
ArXiv CS.CL
Source
Generated insight by OnAirToday AI
This research highlights the critical distinction between an AI model mimicking behavioral patterns and possessing actual human-like cognitive biases. By creating a contamination-aware benchmark, the study provides a robust framework for disentangling training data echoes from emergent psychological phenomena in agents.
Comments
Keep comments relevant and respectful. No links, advertising, self-promotion, or spam.