pretty much what the others have said, especially these days a lot of the advancements have been engineering improvements (better methods for training/finetuning/rlhf, quantization/other optimizations), you want to do a PhD and actually do research you definitely have to like math and theoretical stuff (and tearing your hair out), but the vast majority of actual work is just engineering skills (applying other people's research)
on the topic of research i feel like ppl see the AlphaGo deepmind video and stuff like that and say omg this is the dream!! but if you really think you want to do research you need to accept the fact that you will have to stay up to date reading papers constantly (Pain) and there's a huge potential for burnout, it really isnt as sexy as it sounds (i did my masters and got a job and then realized i hated it, having to sit at a desk for 8 hours reading papers and working on shitty research code / be expected to write a paper on your own in 4 months, shit made me wanna kms)
also phd applications are even more competitive and almost impossible to get in unless you have a first author paper at a top conference, and a large number of ML or even data science jobs require a PhD, it sucks but if there's an exciting field that also pays well, odds are it's going to be very competitive and hard to get into
my main advice for ppl who want to get into cs stuff in general is to get as much real world experience or side projects as possible, the tough thing is if its a PhD people are willing to believe you learned something in your years of research, but a masters student who did only research for two years a) still can't get into phd programs very easily and b) employers don't give a shit about ur research and want real world exp ive seen some of the most horribly written code -- phds who got into top conferences doing their entire project in a jupyter notebook