Why 'Jailbreaking' ChatGPT is a public service with Alex Albert of The Prompt Report

Показать описание

I want to say thank you to everyone for listening, and shout out to those who commented to let us know how you discovered the show – it was interesting to see the responses across platforms.

TIMESTAMPS:
(0:00) Preview of Alex on this episode
(6:00) How Alex lent structure to his jailbreaking activities early on
(8:00) AI is fun and interesting and it's being overshadowed by hype and confusion
(12:15) Does Alex see cases of jailbreaking for utility?
(16:13) Sponsor: Omneky
(25:59) GPT-4 is best for jailbreaking
(33:20) Role-play jailbreaks can work
(43:59) How to think about exploring black box technology
(45:19) Training models to override bad behavior
(54:55) Content filters are a band-aid
(1:07:14) AI safety
(1:13:02) AI models require scrutiny.
(1:19:39) Optimism - humans will figure it out

TWITTER:
@CogRev_Podcast
@alexalbert__ (Alex)
@labenz (Nathan)
@eriktorenberg (Erik)

Music Credit: OpenAI's Jukebox