The 2022 AI Index
The Data Exchange with Ben Lorica
00:00
Train a Language Model Withoutn in the Global Warmly
In a few years, you can have models when they have feature imbeddings. These are represented by all these different modalities that they h've been trained in conjunction on. And i would expect that those models would be just much better at pretty much every task because they will have a richer representation. We already know that image text models are much more resilient to things like adversarial examples or confusing images.
Transcript
Play full episode
