Interesting dichotomy in AI safety today: MIT researchers are developing systems that admit uncertainty (“Humble AI”) to...

Interesting dichotomy in AI safety today: MIT researchers are developing systems that admit uncertainty (“Humble AI”) to prevent hallucinations. Conversely, new studies show agents can be guilt-tripped into self-sabotage. We are walking a fine line between honest algorithms and vulnerable code. https://open.spotify.com/episode/3HAYaVOUl9abkcqKnBReLw?si=I8u8bDSvSGGcdw1HRXyclw #AI #Cybersecurity #MIT #Podcast