If you know about math you're just as good at ML as a person who read everything about swimming is in swimming. You got to run experiments to see what happens, build an intuition, understand the problem from the inside. Math leaves you with a few pretty formulas and nothing else.

Ah, you've almost described the contemporary profession of being a Machine Learning Priest.

There are lots of ideas floating around. Everyone who has studied the field has ideas. Ideas are cheap, results matter. The problem is we don't know if any of these ideas would work, and proving an idea requires lots of data, simulations and compute.

Being good at grasping the theory is just the first step in a thousand mile journey. The problem of AI is not going to be solved with a neat math trick on paper, but with lots of experiments. Nature has taken a similar path towards intelligence.

The field of statistical inference is about mathematically proving how good various statistical ideas are. It's possible to do better than just throw trial and error on a new idea.

> much of ML-math in a day.

ML is not AGI