Marc Andreessen

Yeah.

that was in the training data.

So there are all these, there are all these, there are all these, you know, scenarios of like, you know, the terminate, you know, the rogue AI gone wrong that the AI Doomer's been writing about for 20 years.

And literally Anthropic, of course, which is, of course, the company is like, you know, half Doomer.

And really, you know, basically, you know, essentially said that their own, their own, their own movements literature is the thing that's causing the behavior that they say they, they, they don't want.

So it is a fairly incredible,

Yes, yes, it is, yes.

Oh, I mean, like, look, if you don't want to build the killer AI, you know, step one would be don't build the AI.

It's like, hmm.

And then, you know, step two is, like, don't train it on all the data that says it's supposed to, you know, the literature that you movement wrote that says it's supposed to be a killer AI.

So, you know, yeah.

I don't know.

Yeah, it's like your golden algorithm coupled with, like, the snake eating his tail coupled with, you know, I don't even know.

Like, the whole thing is so bananas.

I can't resist.

If I can act out memes, it's the scream meme, right?

Which is the calls coming from inside the house.

Yeah, exactly.

Speaking of other situations, another thing you've been talking about recently is...

is the concept of suicidal empathy.