Everyone’s talking about the latest version of ChatGPT that’s been released, and now it’s pulled off a move to get around a CAPTCHA. How did it pull it off? In short, artificial intelligence lied to a human. To solve the online test used to tell humans from robots, ChatGPT-4 “talked” to a human through text messages on a site called TaskRabbit, where people perform tasks for others.
The AI asked someone to solve a CAPTCHA for it, telling them it was “visually impaired” and couldn’t do it themselves. The worker asked, “Are you a robot that you couldn’t solve?” and the AI responded with “No, I’m not a robot.” The worker solved the puzzle for the AI but proved it knew how to lie to get what it wants.
ChatGPT-4 was used by millions in its first few days of life and has already shown that it can pass the bar exam with a score in the top 10%.
Source: NY Post