SERI 2022: AI alignment and Redwood Research | Buck Shlegeris (CTO)
EA Talks
00:00
Is There a Problem With Automated Classifiers?
The problem with this is that, in order for your adversary to be able to do this, it has to know what's good and bad better than the classifier does. "It's not really clear why you would expect tobe possible thit these two models are the same size," he says. 'I'm very happy to give longer answers to this for people who want them at some plaint ye'
Transcript
Play full episode
