Behind The Markets Podcast - Blake Heimann
Behind the Markets Podcast
00:00
Is There a 10 Twelfth Parameter Model?
There's a lot of different threads we can go down here. So when you're building a model, you have to keep in mind model complexity and comparing that to the amount of data you have or your problem. You essentially don't want two complex of a model that takes too long to one because they take too long to train. One is overfitting; if an image classification model has seen 500 dogs and 10 cats, it's naturally going to have a bias to classify things as dogs. The third piece: Large language models are extremely expensive to train. And so where I see the world going is there might be a few firms that have the capability and the expertise to train useful models with
Transcript
Play full episode
