Alex Tamkin on Self-Supervised Learning and Large Language Models
The Gradient: Perspectives on AI
00:00
How to Train a Model to Do Two Domains
The contrast of learning based out with Emix does better on continuous domains like audio or images and other domains. One shed that's based more on an LP based objective does better on text. So you see some interesting patterns there. And so we hope to introduce new domains to basically test how general models that people propose are and see whether they do generalize to these more sort of higher impact real-world domains.
Transcript
Play full episode
