Episode 02: Sarah Jane Hong, Latent Space, on neural rendering & research process
Generally Intelligent
00:00
How Do You Think You Can Generate Wolverine?
In the limit, things can be stored in weights of a model. We've shifted our minds more towards a large, suber base model. If you combine this with basically retrieval augmented techniques, you can achieve similar performance on a smaller model,. which means that you have more gains to continue scaling up.
Transcript
Play full episode
