[dragonwriter]

> If no one else does it soon I'll probably do it myself: we're long overdue for the ad-block of LLM output. I want a browser plugin that nukes it at the DOM, and I don't care how many false positives it has.

Well, if you don't care how many false positives it has, just block everything. But there's no even remotely reliable way to detect LLM output if it isn't deliberately watermarked to facilitate that, so you aren't going to get anything that is actually good at that.