Fwiw this feels deeply relevant to my usage of LLMs to structure data. I’d like exactly a good indicator of uncertainty for each bit of data.