Comment by qwerpy

The bots (and humans) I’m trying to weed out are the pot-stirring permanently outraged/bitter commenters. For example on a “car overturned on X road” post, people getting into fights about whether BMW or Subaru or Tesla drivers are worse. That’s all noise. I want to know when it happened, which direction of traffic, if it’s cleared up yet, etc. LLM works well to remove the off-topic garbage. If someone is botting the actual content (recommendations, reviews, etc) and the up/downvoters are unable to identify it as a bot, I probably won’t be able to either. So I’m not worse off using the LLM.

Worst case I get some pre-digested garbage and I waste less time, best case the LLM pulled out a few good nuggets of info. It’s not a perfect system yet. There’s a lot to tune with how deep to go, what constitutes garbage vs signal, and so on. At least it’s fun :)