EP 31 Noam Shazeer - Google veteran and AI inventor on future of AI
The Aarthi and Sriram Show
00:00
How to Build a Probabilistic Model?
As models have gotten bigger and better, the early relatively small models would mostly get the grammar right. But as you start making them bigger and training them more and making them better in various ways, then they start okay being like perfect the grammar but getting a lot of facts wrong. It's maybe it is a lot like a child learning that at the beginning, you know, like if you have like a very, very weak model, it'll put together letters that seem to make sense and dabble or something. And then at some point it'll get the grammar mostly right. As the computing power goes up, so does the amount of training in the size of the model go up.
Transcript
Play full episode
