Alex Tamkin on Self-Supervised Learning and Large Language Models
The Gradient: Perspectives on AI
00:00
The Importance of Pre-Training Objectives
I'm curious basically how that went and then what was involved in actually coming up with something that could tackle this. Yeah, so we come up with some sort of baseline algorithms to get people started. But there's still a whole lot of work for coming up with more general purpose pre-training objectives. Right. Emix and Shed actually do help broadly, but they don't always help evenly and they don't help on every single domain.
Transcript
Play full episode
