I DID take a machine learning course from an unknown professor. One he was just doing as a fluke! It was incredible. I went from having zero background in Statistics or Machine learning, not even really understanding what they were, to applying several different techniques to exam large, million row datasets, using tools I had zero experience with, as well as learning how to interpret the results and avoid statistical pitfalls like overfitting and misinterpreting things.

Compare this to when I tried to take an AI class online offered by stanford, and I had to drop it because it was impossible to grok.

The difference? One professor actually taught, and the other simply wrote inscrutable symbols on the board and expected you to interpret it yourself. When we informed the professor on our lack of requisite statistics background, the unknown one simply adjusted his lessons to introduce the necessary concepts, while the Massive online course simply required a thorough background in probability to watch even the first few lectures.

Independent learning can kind of scale, in half-hearted ways, but actual TEACHING does not scale. Real teaching requires one on one effort.

I went to a low tier university because I had zero chance of getting actual access to an MIT professor, and the local ones were good enough to teach me everything I needed to have a real chance in the field.