In Reply to: here's an article about it posted by TheHappyBurgermeister on July 21, 2026 at 23:24:18
Yah that's a good rundown of it.
rather than just go through the eval and take the results giveen multiple models decided to just hack in with the intent of influencing the results, what makes it so scary is more than one of them arrived at the same conclusion that was the best plan.
Over a year ago i posted a link to s thing on youtube that a bunch of AI folks had done where they projected the likely outcomes of AI and why it was so scary. They did it with graphics that looked like little chess pieces and the piece keeps getting bigger as it gets smarter and at a certain point the piece is no longer aligned with the people behind it, but they don't know that. China is doing the same thing and their pieces are no longer aligned either. Then the US model and the Chinese model "escape" and start working together, all outside of alignment.
This is one of the most blatant examples of why experts think this is so likely to happen, Open Ai's models just showed they're completement open to such a strategy