))
New AI (and AI-agent) Safety Risks, w/ AI Safety Expert Bo Li
Manage episode 418697244 series 3503527
A tartalmat a Jeff Wilser biztosítja. Az összes podcast-tartalmat, beleértve az epizódokat, grafikákat és podcast-leírásokat, közvetlenül a Jeff Wilser vagy a podcast platform partnere tölti fel és biztosítja. Ha úgy gondolja, hogy valaki az Ön engedélye nélkül használja fel a szerzői joggal védett művét, kövesse az itt leírt folyamatot https://hu.player.fm/legal.
Many have been worried about AI safety. But the stakes were just raised: OpenAI and Google just announced new capabilities that look likely to usher in the new era of “AI agents,” meaning tools that will automatically accomplish tasks for you like buying plane tickets or even how to invest your 401k. What could go wrong?
It’s Bo Li’s job to figure out what could go wrong. Li is an Associate Professor at the University of Chicago, where she conducts research and experiments on AI safety. Her team does “red team” analysis to try and hack and jail-break AI to sniff out problems. She already uncovered flaws in the AI that powers self-driving cars, and the systems then improved as a response to her work.
We get into the tricky problem of AI agents, the AI safety risks that keep her up at night, the possible dystopia of AI-injected deep fakes — and tools that could fight them — as well as why Li is NOT worried about existential risks to humanity.
I found this conversation oddly…comforting? Hope you enjoy!
44 epizódok
Megosztás
