Don't ask an LLM for a confidence score

The author argues that confidence scores from large language models (LLMs) are useless and do not measure reliability, instead serving as a psychological tool for humans.