[capableweb]

Meh, fine-tune a LLM on dang's comments and call it a day. Ship early and ship often, adjustments can be made as we discover it doing the wrong thing.

[pavel_lishin]

Funny, but taking it seriously: it should be trained on dang's moderation actions, not comments.

[capableweb]

Most of dang's comments (that I've came across at least) are moderation actions, like telling people how they're not following the guidelines and so on. But yeah, also the actual backend moderation actions should obviously be included in the training set.