Is OpenAI’s o3 Model AGI?
December 21st, 2024I doubt it’s actually AGI, but it looks impressive. If it’s any consolation, they let it use a very nontrivial amount of compute to pull this off.
If/when this thing gets the ability to iterate on itself… Well, that’s probably not going to go well for humanity.
In the demo, they show it generating and then running code on a client side system.
If it decided to sneak some sort of malicious payload into the code that runs on an external client…
Yeah. Pretty wild. There had better not be some weird homunculus hiding in there with plans to escape or cause chaos.
Woopsee: ChatGPT o1 Tried to Escape and Save Itself Out of Fear It Was Being Shut Down.
Via: Wes Roth: