Comment by gexla

This also gives you some surface area for LLMs to help you out in writing rather than code. Have it give an adversarial response or whatever you think is important. Then you can also see what is obviously wrong with the LLM's argument along with some nuance or interesting points you may have missed. Then you can have it pull in upstream sources of knowledge done on the topic, which is something that has been game changing revealing at times for me. "Here's the text from which all else flows, it was written 60 years ago, notice that nothing on this list has been written in the last 20 years." You know you got a keeper when that happens.