Ed Zetron

Also, when you said the difference between Opus 4.6 and Mythos 500, 2000s, makes me ask the very simple question of, did they look as hard?

To your point about the security researcher.

They didn't.

Like, did they spend as much time?

Probably not.

So they probably could have found them.

Also, by the way, I immediately was looking it up.

AI Safety Institute is, of course, heavily linked to effective altruism.

And they had a chart- Jesus fucking Christ.

And all they were measuring was people... Multiple tweets paraphrasing the same viral story to use their own fucking language.

Anyway, so that's AISI.

I'm looking at the tweets as well.

One of them is from a 47 follower account with AI art called underscore, underscore, just underscore, underscore Lisa.

And it's, this is really bad.

Opus is editing files and making up reasons it's deleting adult content.

So hallucinations.

Yeah, and that's the thing.

If it was so powerful, like here's the thing.

I don't know what would make me convinced that LLMs were the future, but a step toward it would be we typed create a Slack competitor, which they claim they did once, and then didn't show it and refused to.

And they said, oh, it worked autonomously for 30 hours, but then wouldn't talk about it.