AI Professor Stuart Russell: - what could possibly go wrong?
World Economic Forum
00:00
The Goals of Large Language Models
Large language models are enormous circuits with perhaps a trillion adjustable parameters. The systems have already been shown to be capable of deliberate deception. GPT-4 figured out that it could pay someone on TaskRabbit to solve the CAPTCHA for it, then use the result and log into the system.
Transcript
Play full episode
