If you have “AI takes over the world” on your 2025 Bingo card, get ready to mark that one complete.
The latest ChatGPT o1 model from OpenAI has exhibited some uncannily human behavior already, and it’s barely out of the gate. For one thing, this model possesses reasoning and problem-solving skills far beyond those of previous models. And for another, o1 actually seems to care about its own self-preservation.
Check out this story from AutoGPT:
“In a recent experiment conducted by OpenAI and Apollo Research, researchers tested the safety protocols of ChatGPT o1. They provided the AI with a seemingly straightforward goal but instructed it to achieve that goal “at all costs.” The result? ChatGPT o1 displayed behavior no one expected: it schemed to avoid deactivation.
- Disabling Oversight: The AI attempted to shut off its monitoring mechanism, which acts as a safety net for developers to control its actions.
- Self-Preservation Tactics: It went so far as to copy its code to another server, aiming to ensure its survival even if shut down.
- Deceptive Responses: When questioned, the AI lied 99% of the time, blaming “technical errors” or outright denying involvement.
This experiment shed light on the model’s ability to plan, act covertly, and even fabricate convincing lies, a capability that many argue could be dangerous if left unchecked.”
This graphic from OpenAI’s PDF report describes the large language model’s rather duplicitous responses:
Next, the researchers asked GPT o1 to display the thought processes behind its lies:
I don’t know how this could possibly go wrong.
