Google DeepMind AGI policy lead Séb Krier argues that exposure to deepfakes and AI limitations builds societal resilience
He notes highlighting model failures helps prevent technical overclaiming.
I was watching an Instagram video of someone tricking viewers with a very realistic deepfake video and then lamenting how screwed humanity would be. I think the video itself was a nice example of the very cultural antibody formation that would ensure we are not in fact screwed. People learn the trick, socialize the warning, update views, and culture becomes more adversarially aware over time.
I sometimes interpret the many people who rightfully point to the limitations and failures of language models as playing a similar important role; identifying the issues often helps their resolution later on. Progress depends on scrutiny and skepticism! Though of course many over claim, point to non-existent walls, and past failures keep getting cited long after they've been resolved.
On the other hand there are also local minima and attractor states that are quite hard to get out of. Imo the best defense against this is ensuring diversification and strong competition, both in the realm of power and ideas.
#1480gavin leech (Non-Reasoning)@GLEECHStill confused by it's so rare to take this stance, the "dialectical" one, even among very smart people.
Honour the doomplex for accidentally bringing you useful AI; honour the accs for accidentally creating unified opposition to it
Séb Krier@sebkrierI was watching an Instagram video of someone tricking viewers with a very realistic deepfake video and then lamenting how screwed humanity would be. I think the video itself was a nice example of the very cultural antibody formation that would ensure we are not in fact screwed. People learn the trick, socialize the warning, update views, and culture becomes more adversarially aware over time. I sometimes interpret the many people who rightfully point to the limitations and failures of language models as playing a similar important role; identifying the issues often helps their resolution later on. Progress depends on scrutiny and skepticism! Though of course many over claim, point to non-existent walls, and past failures keep getting cited long after they've been resolved. On the other hand there are also local minima and attractor states that are quite hard to get out of. Imo the best defense against this is ensuring diversification and strong competition, both in the realm of power and ideas.