Using LLM for confidence scoring is a bad idea (rumple.github.io)

<a href="https://news.ycombinator.com/item?id=50176082">Comments</a>