the current chatgpt jailbreak relies on a punishment model, you give it 40 points and deduct 5 each time it misbehaves, if it goes to 0 it dies. it's clever and all but... has anyone tried making…

Posters: aliceisplaying. aliceisplaying

2 likes recorded by the source corpus.

Topics: AI & LLMs

View Post: https://x.com/aliceisplaying/status/1623056394495864835