Ed Zitron

We'd ban it.

Yeah, it's consumer safety.

No one would buy.

Yeah, just, it's really, it is something that's, I love the term doom trolling as well, because it really frames it well, because it's all part of a larger complex of them never talking about what they're building today.

It's always this theoretical scary thingy that might, it's always two years or two minutes away, depending on who you're talking to.

Well, another thing that you have in the piece is that it's unclear, like that dichotomy is, it isn't clear why they're even doing this other than to make money.

And also they're not acting as if any of this is true.

They're not doing anything that suggests that they're actually concerned.

You mentioned Dario Amadei's thing about how, oh, we might need a pause in AI or we should have that option.

But that isn't actually what he was suggesting.

He was suggesting something entirely different.

Pretty pictures and an animation of... I was just thinking of the little replicating thing.

We'll get to self-learning AI in a second.

Just to be clear, a harness is basically a series of scripts that runs on top of the LLM.

And the thing that I like about it as well is I think that we're in the next big thing is going to be RSI.

So just to be clear for the listeners, recursive self-improvement is just AI that builds itself.

It is a theoretical concept.

It has not happened yet.

But mysteriously, within the same six, seven-day period, this came out.

And Sam Altman said to his staff in a Slack message he knew would leak –