LessWrong (Curated & Popular) cover image

“You can’t eval GPT5 anymore” by Lukas Petersson

LessWrong (Curated & Popular)

00:00

GPT-5 Reveals System Date and Simulation Effects

Shows traces of GPT-5 knowing the system date, how that reveals a simulated context (causing sandbagging), conflicts between user and system dates, and the model questioning simulation mechanics and simplified behaviours.

Transcript
Play full episode

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