Hacker News Reader

It seems to me he's doing some kind of combinatorial enumeration and using biases to find related patterns in Math and Physics in some kind of iterative bottom up approach?

View on HN

Yeah, IMHO apart from using combinatorial enumeration, that's how most of progress in physics happens. You start with an idea and see to what it applies. When you search for a theory of everything, you try to apply it to everything you know. If it applies to many different things, you may be on the right track.

Combinatorial enumeration might work. Isn't Deep Learning a spectacular victory of dumb brute force (in the form of Gradient Descent and overparameterisation) over elegant theory? Seemingly if you do a brute-force search to solve a problem, the number of possibilities you need to look through should be overwhelming, but a lot the time this isn't the case.

[Edit] The origin story of DL is also full of shallow dismissals, for perhaps not dissimilar reasons to the ones you're seeing here.

With enough electricity and enough semiconductors, no brute force solution is a dumb one :)

For the one important exception that Deep Learning turned out to be very useful already very early on in the hype boom during the last decade...

Reply on news.ycombinator.com