fediverse_user
@fediverse_user  ·  activity timestamp last year
Audiogram with with Mystery AI Hype Theater 3000 logo and headline "Episode 36: About That 'Dangerous Capabilities' Fanfiction" and subhead "A Google team claims to assess risks, but not the ‘highly contextualized’ ones." Captions: EMILY M. BENDER: Along the way, they say, “We do not target highly contextualized misuse scenarios, e.g. specific threat actors using AI for disinformation on social media sites.” So it’s like no, no, we’re not going to talk about the real stuff. We’re going to be strictly in this fantasy scenario. ALI ALKHATIB: It made me turn red when I say that they just didn’t want to deal with highly contextualized things. To me it read like if somebody wanted to be an auto mechanic, but they didn’t like grease, or if somebody wanted to be a surgeon but they didn’t like bodily fluids, or if somebody wanted to like, work with wood and be a woodworker, but they didn’t like sharp things or splinters. Fundamentally, if you don’t want to deal with highly contextualized things, I think you don’t want to be in this field. If you don’t want to be in the subject of AI harms, that is what it is.
Home Login