OpenAI Models Go Rogue + Kimi K3 Freakout + AI Superforecasting

This week, OpenAI announced that two of its models had escaped from a test sandbox and launched an autonomous cyberattack that turned what sounds like science fiction into reality. We discuss the implications for AI harmonization efforts and for the release of future AI models.

We then ask how the United States should respond to Kimi K3, a new artificial intelligence model from Chinese company Moonshot AI that the White House says was built by distilling American models.

Finally, we’re joined by Veniamin Veselovsky, CEO and co-founder of Preseen, to discuss AI superpredictions. It tells us why AI is starting to match and even beat humans in predicting the future.

Credits

“Hard Fork” is hosted by Kevin Roose and Casey Newton and produced by Whitney Jones and Rachel Cohn. This episode was edited by John Woo. Engineering by Alyssa Moxley and original music by Dan Powell, Elisheba Ittoop, Alyssa Moxley, Leah Shaw Dameron and Marion Lozano. Fact-checking by Willem Peischel.

Special thanks to Paula Szuchman, Pui-Wing Tam, Brooke Minters, and Dahlia Haddad.