The second document was a long, quite boring, government document. It was so long, pasting it into the chat window crashed the UI. (118 pages as printable PDF, but I was pasting in the plain text.)

I found what I was looking for the old fashion way—briefly skim-reading the document, then searching for the relevant keywords.

I tried to get a large language model to summarise some documents today.

The first one was a transcript of a podcast. It didn't pick out anything interesting - merely stating the topic of the podcast (which one could infer easily from the name of said podcast and the professions and backgrounds of the guests).

@Edent the inevitable end of a bad idea