Measuring Behavioral Drift in LLMs: 22 Signals, 5 Dimensions, and the Calcification Effect
中文摘要
借用行为科学工具量化大型语言模型性格漂移，发现“钙化效应”。
English Summary
Researchers developed a framework using behavioral science to quantify personality shifts and behavioral drift in LLMs through 22 signals and five dimensions, identifying the "calcification effect."
原文节选
How we borrowed LIWC, OCEAN, and VAD from behavioral science to build a reproducible framework for quantifying personality shift in AI… Continue reading on Towards AI »