The GPT-3 Architecture, on a Napkin
Daniel Dugas,
Artificial Curiosity,
Jul 31, 2020
I've been writing about GPT-3 fairly often over the last couple of weeks. I've mostly highlighted what it can do. If you're interested in how it can do what it does, then this accessible article is for you. It helps to have a bit of background in machine learning (ML), but honestly, you won't need it. Where this article is really useful is in helping you conceptualize what GPT-3 is doing under the hood - you won't be an expert but you'll be able to talk knowledgably about the type of processing it does. There's a lot to it - but at heart, it really boils down to counting strings of symbols and predicting which will come next.
Today: 1 Total: 36 [Direct link] [Share]
