TAME Pain

LLMs struggle with confident fabrication in pain detection tasks

A new study evaluated six large language models on their ability to distinguish between actual pain signals and fabricated ones in clinical speech transcripts. While most models correctly abstained from predicting pain …