I started my undergraduate studies at UIUC wanting to eventually pursue a PhD in math and become a math researcher. For the first two years, I was in what I call “math mode.” After my second year, when I took two AI classes in one semester and did a summer internship applying AI to gravitational wave detection, I switched to what I now think of as the “deep learning mindset”. I further adopted it as I pursued my MS in AI at Stanford.

In the math mindset, my thinking was oriented around proofs, correctness, and precision. I would refuse to engage with something that was not well-defined, since in math the onus of definition falls on the person making the claim, not the listener. Nothing mattered until it was rigorously proved, et cetera.

In the deep learning mindset, things are very different. Everything is hand-waved away. Empirical results matter far more than theory. No one cares about proof; people just try different things- architectures, datasets, loss functions, learning paradigms- until something works. If something works better, you adopt it without worrying about proving its optimality, well-definedness, or statistical guarantees.

Initially, I was quite annoyed by this. But I came to embrace it, and now find both mindsets useful in daily life.

The deep learning mindset teaches you that when you are unsure of something, instead of falling prey to the tempting solution of hand-engineering heuristics, you should focus on getting your initial conditions, learning signal, and architecture right.

This often applies to teams. In any group project- a startup or research lab, say- if your initial team is good and motivated (great init, good architecture), and you listen to your empirical results (market feedback, research outcomes), move fast (good learning rate), and backpropagate signals into the team well, you will win.

In this situation, the math mindset leads to analysis paralysis, and does not serve you well here.

This mindset shift extends to other fields too. Programming has its own mindset, as do physics, manufacturing and other fields. Being able to pick up these mindsets and decide which model of thought applies to a given situation seems like a pretty valuable skill.