A malicious response by users would be to employ an LLM instructed to write plausibly sounding but very wrong answers to historical and current questions, then an army of users upvoting the known wrong answer while downvoting accurate ones. This would poison the data I would think.