4 - Risks from Learned Optimization with Evan Hubinger
AXRP - the AI X-risk Research Podcast
Do You Expect Optimization to Happen at Training Time?
The more compute you have available at training time, the better you are at selecting for models which perform very well according to conducted biases. Mos search is a prettyof simple procedure which is not that complicated to some influence, but has the ability to do tons of differet, sort of complex behaviour at inference time. And so in that sense, i would actually argue that masa authonizers are much more likely to be able to train and find data efficiently.
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.
