AXRP - the AI X-risk Research Podcast cover image

17 - Training for Very High Reliability with Daniel Ziegler

AXRP - the AI X-risk Research Podcast

CHAPTER

What Does Quality Mean?

The paper uses an ensomble of networks, each one generated by taking a pre trained model. They replace the final layer with a ranimly initialized linear layer and fine tuning the whole model based on human preference comparisons. I think maybe the upshot was ansamling does work, but it's expensive because you have to have a bunch of copies your model. And in some sense, i think it's like, sort of ok, or, like, its sort f doesn't matter exactly what metric o use here for this point of our research, as long as we are being consistent with it.

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.
App store bannerPlay store banner