Comment by trollbridge

Well, you can use LLMs to parse LLM-generated slop. They make nice summaries. I have taken this approach to people who send me obviously generated LLM text; I simply run it through an LLM, paste the summary, and ask them "Is this an accurate summary?" and then I ask the for their original prompt.