20 - 'Reform' AI Alignment with Scott Aaronson
AXRP - the AI X-risk Research Podcast
00:00
Artificial Intelligence Alignment: The Off Switch
Computer scientists had been thinking about backdoors as, you know, in AI safety failure. But they could also view the insertion of undetectable backdoors as a positive for AI safety. And what's interesting is that from the evidence at all, it may be possible to insert a cryptographic backdoor into your powerful AI. So now we consider what person or company might have inserted and are trying to get rid of?
Transcript
Play full episode
