All-In with Chamath, Jason, Sacks & Friedberg cover image

E116: Toxic out-of-control trains, regulators, and AI

All-In with Chamath, Jason, Sacks & Friedberg

00:00

Open AI's Trick to Fool the Trusted Safety Layer

Reddit users figured out a way to hack the AI's trusted safety layer,./nIf they didn't give the AI the answer to a question, they would lose tokens.

Play episode from 01:18:02
Transcript

The AI-powered Podcast Player

Save insights by tapping your headphones, chat with episodes, discover the best highlights - and more!
App store bannerPlay store banner
Get the app