It looks like it's going to take many, many hours of reading and thinking to understand GPT-4 and SD the way I want to.

Interesting meta on the way I read and rabbit-hole on subjects. I did a lot of this in the early COVID days, probably a few hundred hours of reading. I'll start with one or a few main topics, and in the first texts I'll run across several things I don't understand, which causes a branch where one source document may have me reading 5 more, which then lead to 20 tabs open, etc.

I get a sense of how big a subject is going to be by how many branches deep I am before that rate of new branch opening starts to slow...and then eventually drop below 1 so the total active branches start to decline.

There is SO MUCH that I don't know about neural net implementations and large language models that I can tell this is going to take a long time. I've made a tiny bit of progress in four or five hours, but not even a hint of where the end of the initial exponential phase will be.

Kind of fun, kind of frustrating, kind of exciting, kind of overwhelming. With my COVID reading, I finally just punted on really understanding the immune system as a whole because that would have made the reading list like 5x longer...

And as with COVID, OMG is Wikipedia a wonderful place to start each major branch. It's like it was designed for this sort of learning.