23 - Mechanistic Anomaly Detection with Mark Xu
AXRP - the AI X-risk Research Podcast
00:00
The Different Mechanisms That Explain Noise and Sensor One
The way in which these kind of different mechanisms are used is about means and variances and co-variances. Maybe another way to say it is like there was an assumption you could have made during training, which is that noises are never both on simultaneously. And so anytime you have this mismatch, we can like drop something out that only hinders your average prediction by like one part in 1000. But if the things that are happening between train time and test time are quite different, then you should be kind of suspicious.
Transcript
Play full episode
