@Jain @mian they do indeed just grep the transcript, at minimum.

robots that understand nuance are pretty high level things, i can barely get anyone to listen to me about neural search and how easy it actually is. (though enforcing narratives is extremely dystopic, like "you can only talk about X as long as its how its bad")

but, the system doesn't care. per kazynsky (the systems neatest trick) the goal is line go up. if something is controvertial at all that could make line go down, so you are only allowed to do homogenized things that ensure line go up.