You can find the book here. A PDF version is 224 pages long.
I will post progress updates here:
13/04 - P15
14/04 - P16
I will do most of the exercises and some of the problems. This is what the author recommends.
Resolution details
If OpenAI's next SOTA language model isn't called GPT5, but a model called GPT5 is released by them some time later, I will wait until then. If they change naming conventions and stop calling their models GPTn, I will resolve once a new model is released, that is to GPT4, as GPT4 is to GPT3 (or somewhere in that range).
For example:
Their next model is called GPT4V and is much better at visual tasks, and then some time later GPT5 is released; it resolves at that later point.
Their next powerful language model is called ITA1 (Improved Transformer Architecture 1) and it seems like they will never create a model called GPT5, it will resolve for the release of ITA1.
(Let's just hope they call the next model GPT5)
@tr I like it so far. Especially the more technical explanations and exercises, I found the coding parts somewhat confusing, but this could also just be because I know more maths than coding.