20 - 'Reform' AI Alignment with Scott Aaronson
AXRP - the AI X-risk Research Podcast
The Importance of Recognizing Superhuman Performance
Is there a domain where both we could recognize superhuman performance and sufficiently superhuman performance, you know, would set off an alarm that this is dangerous? I think math might be a little bit closer to that. We're noticeably getting better scores on this metric. And we could still score the metric even at a superhuman level. But then there are, you know,. even higher scores on this metrics that we would regard as dangerous. You can also love, imagine tests of hacking ability, deception ability, things like that. It's not clear if anyone in business or government will get on board with such extrapolation.
00:00
Transcript
Play full episode
Remember Everything You Learn from Podcasts
Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.
