>>13539129The link will take you 4-8 months putting in 10 hours a day. I wouldn't bother learning some topics unless you're going to use them. For example, if you wanted to make games, learn Unreal Engine, Godot or Unity, not how to build a game engine. If you want to make game engines then study the components that go into that and how to make them better. Never think that you can complete learning anything either. There is always more to learn. Always question what you know.
>>13539513First of all, if you don't have a PhD you have almost zero chance of ever becoming an AI researcher unless you hit a home run with something and a company like Google, Nvidia or OpenAI notices you. Second, researchers who dedicate every hour of their lives to AI have little chance of getting their papers published or noticed because there are too many papers. You'll need to decide whether you want to be a researcher or an engineer. You won't have time to do both.
You should understand MLPs, CNNs, RNNs, autoregressive and autoencoding transformers, GLUs and RL well enough to create them from scratch since they're used in practically everything. Understanding how batch size and the momentum in ADAM affects training will save you a lot of time. If you only have one GPU, understanding gradient accumulation is a must. If you have a large computer cluster bottlenecked by the network, look into gradient compression. The rest you'll discover and learn from reading papers, not books. Anything older than a year or six months is probably outdated and already improved upon. Once you get comfortable reading papers you should practice implementing them yourself.
When you feel ready to take on the world find some programmers doing what you wanna do and ask them for help. Contribute to their projects and help them with something so they know you're not a wishful thinking beginner. Don't bother researchers though. They're too busy.
And don't forget to save and organize your papers.
