Would be an interesting experiment. My hypothesis is that an LLM would actually do a pretty good job at moderation, but only if the community didn't know it wa

Would be an interesting experiment. My hypothesis is that an LLM would actually do a pretty good job at moderation, but only if the community didn't know it was an LLM doing the work. Moderation is really just applied sentiment analysis, which is an area LLMs are quite strong in. However, as soon as the community learned it was an LLM doing the moderation, it would be torn to pieces by people abusing it. On top of hallucinations, it's one of the major problems preventing many LLM usecases going into production; They are inherently vulnerable to hostile input.