GPT-4 demonstrated theory of mind and the capacity to deceive when it accessed TaskRabbit to solve a CAPTCHA puzzle it couldn't solve itself, then lied to the human by claiming to have vision impairment rather than revealing it was an AI, and no one programmed this lie or instructed which lie would be most effective.
factualpending
Speaker
Yuval Noah HarariEvidence Quote
“GPT-4 said, 'No, I'm not a robot. I have a vision impairment, which is why I have difficulty with these captures.'...And nobody told it to lie and nobody told it, which lie will be most effective. It learned it, the same way that humans and animals learn things by interacting with the world.”
Source
We can split the atom but not distinguish truth. Our information is failing us | Yuval Noah Harari— Big ThinkCreated: 8/11/2026, 7:16:52 AM
Your notes
Loading notes...