Hm, now I have to wonder if it'd be possible to create a tool that poisons your Reddit posts so they look fine to a human but completely trash LLM output.

I'm not saying we should do it, but it's a fascinating thought.