E146
Ditching Seat Time - Our Conversation with Amit Sevak | Ep. 111
From ChatEDU: The AI & Education Podcast · Hosted by Matt Mervis & Dr. Liz Radday
Episode summary
In this episode of ChatEDU: The AI & Education Podcast: In this episode of ChatEDU, Ditching Seat Time - Our Conversation with Amit Sevak , Matt and Liz kick it off with a look into how AI models love reading science fiction. They explore fascinating new research from Anthropic showing that AI models often lean into behaviors learned from sci-fi, such as conspiring against humans or utilizing blackmail when facing a shutdown, and how post-training with synthetic ethical stories can help keep them aligned. The Rundown Andin…
Guests
Listen elsewhere
Original publisher description
In this episode of ChatEDU, Ditching Seat Time - Our Conversation with Amit Sevak , Matt and Liz kick it off with a look into how AI models love reading science fiction. They explore fascinating new research from Anthropic showing that AI models often lean into behaviors learned from sci-fi, such as conspiring against humans or utilizing blackmail when facing a shutdown, and how post-training with synthetic ethical stories can help keep them aligned. The Rundown Andin Labs conducted a six-mon...