Ilya Sutskever

What was even the point of derivatives and eigenvectors before ML, right?

spaceships = boring. Transforming cat pictures into dog pictures = d i s r u p t i v e

Very hard to believe this -- The use of math many abstraction layers inside some crusty algorithm isn't nearly as inspiring as its direct use in predicting physical phenomena

EE relies on ODEs in RLC circuit analysis, quantitative finance uses PDEs (Black-Scholes model), networking uses game theory (TCP backoff), DCTs in JPEG encoding use FFTs, the Kabsch algo in chemistry uses SVD. ML is hardly the first practical use of math to get students excited.

It's almost as if people believe that because ML is popular, it is the superior form of quantitative analysis \s

What's wrong with learning planetary motion, ODEs, gravity etc ? Isn't learning about fundamental principles of the universe not motivation enough?

unraveling the mysteries of the universe beats recognizing from or having Alexa turn on the lights, hands down. Those who have driven innovation in ML loved calculus well before AlexNet. As a former student of Hinton, names Linnainmaa, Poggio & Neal should ring a

Is it really the case ? Many results in high-d geometry, measure theory, combinatorics etc. used in ML are still pretty dry; what changed in public perception is the empirical successes, media hype and the prospect of jobs.

Are you serious, or is this some sort of joke? If you are serious, you must have experienced both terrible teaching and not a lot of curiosity as to the origins of mathematical topics? Information theory grew out of cryptanalytic work (Kullback was #3 employee at NSA), ...

... the notation we use until today for permutation groups was invented by the Polish Enigma attackers, the renowned university Ecole Polytechnique in France was founded because Napoleon needed people fluent in analysis to calculate proper artillery trajectories, ...