As an undergrad, should I study ML or is the field too saturated?
At my school, I've the opportunity to pick the Intelligence thread, I'm really passionate about it. But after coming across threads here about the great stagnation and Deep Learning jobs collapsing, I have to ask is investing money & time to study this field worth it anymore?
IMNSHO, as an undergrad, regardless of the field you end up choosing, you'd benefit from a solid introduction to axiomatic probability, a decent intro inferential stats class that actually uses your foundational understanding of probability, a decent linear algebra class (so you know what an eigenvalue is and how PCA works, at the very least), and a decent class in mathematical analysis.
That will give you the foundational knowledge to study any optimal decision making topic whether it is labeled as ML or AI as such.
Most undergrad ML syllabi I have seen read like bad intro stats courses anyway. However, it never hurts to know enough terminology so that you avoid being overly impressed by snake oil salesmen later (and ML has more than its fair share of those).
In your case that track is just an extra class or two as part of a CS degree, right? If that’s the case, it really doesn’t matter either way. Take it if you’re interested (and you seem to be). On graduation you won’t be distinguishable from someone who didn’t take the extra classes so it neither hurts you or helps you that much. And it’s not like you would say “oh there are no Calc II jobs out there and the compiler job market is very small, let me skip those classes” - that would be ridiculous, those classes just expose you to a bunch of ideas and concepts you haven’t seen before. You can view the ML classes the same way.
Learning a bit of ML is always useful. At worst, you'll get a bit of programming and math practice.
Be a little thoughtful about trying to do ML work full time (there are a lot of people competing for a small number of positions), and be very very thoughtful about describing yourself as an expert after a couple classes.
But go for it! Learning a bit will help you no matter what you decide to do eventually.
I've been cultivating my own theory on that subject: it might be that the field is not saturated, we've just started seeing large scale public successes come out, e.g. AlphaFold, GPT-3. But there is a big chance what is being thought is more operational and probably won't last in it's current form.
Really interested to hear others opinions on that one :-)
If you're really passionate about it, go for it. I'd say we're at a really early stage where the obvious uses are heavily researched, but there will be more niche less funded applications as time passes, like geology. It's like app development in 2010.
I'd say go for general software dev. If you still want to do it, get a masters or PhD in it. Most places won't let you build models without a higher degree in it, so you would just end up being a data jockey for the people who do build the models.