Hi Matt
This gap definitely also exists in ML. Even within research, there is a huge gap between the theoretical papers, which usually have to make ridiculous assumptions, and the practical papers. Even those practical papers then are mostly irrelevant for business because most don’t work. They find some .1% improvement on some benchmark, but it’s not robust at all, or flat-out fake. ML has a redeeming quality though, since there are a lot of industry-players in research (Google, Meta, etc.) and that usually leads to results that can be applied in industry, so I guess in total it’s somewhere in the middle in terms of how bad the situation is.
As for having a deep understanding, I generally am pro understanding things because I think it leads to better diagnoses of problems and more creative solutions, but I see many many people in industry just hacking around bit by bit and solving most of the problems to a sufficient degree, maybe even better because they’re usually more pragmatic, so I wouldn’t say it’s a necessity.
1 Like