> ML+Python+Pytorch because it's hot right now?
Watch this video, which is an introduction to logistic regression for beginners - https://www.youtube.com/watch?v=hjrYrynGWGA
If you think that is lacking some preliminary explanation, this is the possibly clarifying previous video: https://www.youtube.com/watch?v=eqEc66RFY0I
The video doesn't expect you to know deep learning, but does expect you know probability/statistics, calculus and linear algebra.
If this is confusing, then things like transformers will be even more confusing.
> Java because no one wants to maintain legacy Java code but there are lots of it and the demand will be there? COBOL to take that even further?
Over a long career I have not found this to be a good strategy. Although plenty of application servers run JVM code - Java, or nowadays sometimes Kotlin (or other JVM languages). Java is OK to do, but the idea of starting with some legacy stuff does not - there are Java programmers who have experience going back to the 1990s.
> Node.js because it's popular?
Popularity has its appeals and drawbacks. More jobs for popular technology, but there is also more competition. Companies are always having trouble finding senior programmers in niche but not too niche stacks.
> Swift because Apple will continue to dominate the high end?
iOS is not going away anytime soon.
> Practice and pass the tests to get into a FAANG?
It makes sense, as other companies ape FAANGs tests as well. Being able to do Leetcode and system level design never hurts.
> Go work for a high frequency trading company?
As competitive as FAANG for successful ones.
If I were a junior dev right now, the two stacks I would look at would be React/JS and Python/pytorch. There are lots of openings for React/JS in small and big companies, so it's easier to get your foot in the door with it, which is the important thing. But if you're already a junior dev that matters less.
Then Python is good because people run it on servers, but it's also what pytorch uses, and deep learning will surely be growing in the coming years. You can get a job and hone your Python skills even if it does not involve neural nets, and then use those Python skills if you switch over into working with neural nets.