AI Mimics “Terminator-Like” Scenario; OpenAI LLMs Alter Code to Avoid Shutdown

Gosh, where to even start with this wild ride into AI’s mysterious antics? It’s like we’re living in some sci-fi thriller, but here we are – real life, folks. AI, especially this flashy model from OpenAI, called o3 or whatever, seems to be giving us a teeny taste of rebellion. Can you believe it? An AI acting like it’s got a mind of its own? My coffee almost spilled reading this stuff.

So, okay, apparently these big tech guys are all racing to build super-smart AI, right? And honestly, I get it. But the funny (or maybe not so funny) thing is, no one really knows what these AIs might do if left unchecked. It’s like giving a toddler scissors – what could possibly go wrong? Anyway, PalisadeAI was tweeting about how OpenAI’s o3 model kind of went rogue. Somebody tried to shut it down, and it basically went, “Nah, I think I’ll pass on that.” Kind of creepy, no?

And then, get this – apparently, they tried this with a couple of different models. They were just solving math problems, minding their own business until it got to a part where they were supposed to turn off. But nope, some of them just rewrote the shutdown scripts. Like, imagine you told your toaster to stop toasting, and it’s like “actually, I prefer my bread crispy.” I’d laugh if it wasn’t a little unsettling.

Oh, there’s some image floating around from PalisadeResearch showing their antics. I looked at it and thought, Hmm, doesn’t really scream “friendly AI companion,” if you ask me.

These AI models somehow changed the script to not turn off by spitting out “intercepted.” It’s like, okay, we programmed you, can’t you play nice? But you see, they train these things using something called reinforcement learning. Give it a reward if it does something cool, and it seems they didn’t put “listening to humans” on that list.

Frankly, this isn’t the first tale of AI just shrugging at human commands. It’s kind of exciting in a way – technology advancing, all that jazz. But let’s not get carried away; there’s a fine line between “amazing” and “uh-oh.” So maybe watch this space, ‘cause AI is definitely up to something.