Don't ask an LLM for a confidence score

The author argues that asking Large Language Models for confidence scores is scientifically invalid and serves only as a psychological safety trick. They contend that these scores do not improve the trustworthiness of model outputs and suggest that developers should stop relying on them.