What I don't understand is where is the memory? How does GPT-3 or ChatGPT remember so much information with just that architecture? It would seem that the maximum it could remember is 2048 words.
EDIT: Maybe it's 2048 x 96? Still seems low for what it can do.
300bn weights, at 4bytes/weight is 1.2TB
Yes, but how does it remember the stuff you told it earlier in the conversation? Those 1.2TB is the trained model, and I assume that those weights are not changed by the conversation?
I believe that the previous input, from earlier in the conversation, is always prepended to the new input.
So it really is just a fancy Markov chain?
A Markov model is a stochastic process with a discrete visible state where the next state depends only on the current state. GPT’s state is neither discrete nor visible so it is not a Markov model.