Paresh Devay

And I think that's the...

helpful thing than what makes it similar to spam and why YouTube is comparing it to spam.

But I would also say that detecting AI slop is probably easier than detecting sort of deep fakes and cheap fakes.

There's just something that just looks crappy about it

Yeah.

Another example has been these videos that are seemingly from talk shows, but they're like clips, and then there's these voiceovers that are completely unrelated to the clips that my colleagues reported on last year.

And they're meant to just like kind of get outrage in people because supposedly there's something really controversial happening on this talk show.

But in reality, it's just a real...

sort of clip mixed with a voiceover that has nothing to do with it.

And it's easy, I think, for AI these days to understand that there's some sort of gap here that makes this slop.

But finding a fake is still pretty difficult.

But those can be entertaining.

And if they're labeled as sort of, you know, AI generated, YouTube is okay with some of that content, you know, as long as it's not deliberately meant to sort of be low quality and sort of clickbaity.

Like some people are genuinely interested and entertained by that kind of content.

I would say, you know, you mentioned like

How are they going to do this besides some of the existing algorithms?

I think one place to look at is Kaji, a small search engine, has been asking users to report AI slop so that they can build up a database of AI slop content and then hopefully train an algorithm on that database.

So YouTube might be asking more of their users for...

you know, help identifying slop.

They have a large community of raters who rate videos who could be sort of helping them understand what humans consider to be slop and what they don't consider to be slop.