Doctor Mike

Right.

Like, that's just not

those that that form of AI is not used at all anymore.

Generative AI, um, diffusion models produce much more realistic faces.

Um, they are, you know, you can, the very plastic looking, um, diffusion, you know, faces generated with diffusion models like mid journey and things two years ago,

we've moved two years out now, they're increasingly indistinguishable.

You can actually use, they're called lauras, sort of add-ons that will actually add blemishes to the skin now.

So they don't have that doll-like kind of hyper-realist look anymore.

They look indistinguishable.

And so we've hit a point now where I don't trust my eyes when I look at these things.

It's kind of hard to say to somebody, look, when we're looking at a network of accounts that we think are inauthentic, the way that we identify them is through a series of analyses using AI models to try to detect whether the image is AI.

Can it see itself in itself?

Yeah.

And sometimes they return false negatives.

I don't know if you noticed this, but Grok was asked to identify, to determine whether or not an image of soldiers sleeping in California that Gavin Newsom posted to Twitter was AI generated.

And it said that it was, and it was not.

Right.

There was a moment around October 7th where the Israeli government posted some images of bodies, right, that were upsetting and they were partially blurred.

And Ben Shapiro posted this and the somebody ran it through one of these AI identifiers and returned to false negative.

It said it was AI generated because of the pixelated blur in one portion.