#aisafetytalks — Public Fediverse posts
Live and recent posts from across the Fediverse tagged #aisafetytalks, aggregated by home.social.
-
A lot of focus on controlling open source AI, but how? Seems very unrealistic. Reminiscent of wishful thinking to put backdoors in end-to-end encryption - Stuart Russell #AISafetyTalks #AISummitTalks
-
Large language models are inherently unsafe, because we have trained them that way [LLMs like ChatGPT]. Needs to be designed to be safe up front by design - Stuart Russell, UC Berkeley #AISafetyTalks #AISummitTalks