AI Models Are Being Brainwashed

Hinzugefügt: 2026-06-01

[00:00:00]Anthropic comes out with a new model, they have a bunch of blog posts and here's how we did our safety alignment and very much so there's like more digital ink being spilled of them saying like it knows when it's being tested and it's definitely changing its behavior when it's being tested to get to the point where you aren't doing the testing or you aren't giving it here's the raw base model now be aligned toward what we desire on top of that cuz there's this reinforcement process that you know I'm imagining like a blacksmith with the big glowing red.

[00:00:31]>> There's a brainwashing process. The models are brainwashed.

[00:00:33]>> The AI companies are saying that these broadly are going to be coding themselves in the next year or two.

[00:00:40]>> Yes, >> code. They keep saying, "We're dog fooding it. We're having Claude make Claude." Supposedly, much of or most of the code of these things is already going to be AI coded very soon.