The Future of Language Models

The idea here is how can you use, how can you collect data in our domains where it's even hard to get humans to produce high quality labels or high quality demonstrations. How do we use model plus human together to create higher quality data than a human would be able to create and supervise models that are really hard? I think this is an important problem and probably it'll become more popular in the ML research world. And there might be some fields like biology that are just too complicated for humans. So it could be that we can have them sift through a lot of complicated data from biology and figure something out. Maybe, maybe we'll both be working on it in the future.