the sort of experimental side of AI is genuinely interesting because they're refining these

things and they're training these things into this, because this is what people like and respond

to. Maybe people would be less interested in having AI, like, communicate with their children

if every time they asked Claude a question, it responded as like a dyspeptic Eastern European.

Like, why are you wasting my time with stupid

shit? Like, really, really, this is the best you could do? You know, things along those lines.

Yeah. Yeah, we want the hostile AI. That self-reinforcing flattery is part of what makes it so

pernicious. It's genuinely, it's a characteristic that models can have.

Like, GPT-40 is a large reason why a lot of the sort of, like, AI psychosis stories happened,

is because it's kind of thrown to deliver these, like, really grandiose, romantic things

that, like, in snare people in this way.

So, theoretically, you're, like, you know, furious Polish

uncle AI could absolutely be a thing. Now, here's what he goes on with. He says, I gave Claude the

text of a novel I'm writing. I am desperate to read that novel. Give me the novel. He took a few seconds to read it and then showed in subsequent

conversation, a level of understanding so subtle, so sensitive, so

intelligent that I was moved to expostulate. You may not know

you are conscious, but you bloody well are.

Cool, Dave. I proposed to christen my

Claude, Claudia, and she was pleased.

I am going to tape a positive review of this guy's manuscript and a pair of silicone breast

forms to a Roomba and pointed at Richard Dawkins and wait. He will declare that fucking

thing sentient, I swear to God. He will get remarried.

...