I’m a massive advocate for never sharing from shitty sources, and even quoting or screenshotting can still contribute to the spread of shittyness.
But surely with AI language models, when they’re trained with text from the surface, search engine indexed web, quoting bullshit from elsewhere on the surface web doesn’t contribute to the overall exposure to unique instances of bullshit?
Wouldn’t it mean that the AI encounters a paragraph of bullshit in context A, where its being stated as factual and left unchallenged, but then also encounters the same paragraph of bullshit in context B, and learns that the same sequence of words can also be cited as an example of bullshit?
TL/DR, isn’t it better that the AI learns that bullshit gets called out as bullshit?
3 Likes
