Machine learning is one obvious avenue to go down, yes. Given the relative immaturity of the tech it's actually very possible to become competitive and employable in that area quite quickly, as you're not competing against people with 20 years experience....
Very true as far as the recent advances in diffusion models and LLMs are concerned, someone super enthusiastic about them can no doubt become quite valuable quite quickly. (Machine learning obviously has been around for far longer and there certainly are people with 20 years+ of experience in some fields).
Similarly, I remember around 7 years ago a friend had got a Data Science/ML job at a consultancy, TensorFlow was fairly new and deep learning had a lot of buzz around it, he was working with a team of mostly PhDs and I'd asked him if he was doing much deep learning - he explained that he was but the PhD guys were mostly sticking to quite basic stuff and using Keras as a wrapper and they were a bit shy about writing TensorFlow directly (early versions were a bit awkward at times) - he was suddenly quite a valuable team member as he was familiar with TF and very good at using it, the more experienced guys had done their PhDs in the early 00s when SVMs etc.. were all the rage. A similar sort of situation could well apply again (and certainly did two years ago) re: LLMs, agents making use of LLMs etc.
There's a level of maths needed in data science/ML/quantum that if it's not present becomes a liability and a serious risk (ie businesses making decisions based on data that has not been characterised or even understanding the distribution bias etc).
Very true re: data science, especially w.r.t to the developers who have decided to rebrand as data scientists - about 10-15 years ago there was plenty of hype around "big data" (don't hear that buzzword so much now) and Data Scientist would be the "sexiest career" of the 21st century etc. but if it's some guy with a CS undergrad who confidently chucks data at whatever model he's using from scikit-learn but doesn't really have much stats background then that's potentially quite iffy.
Though in general, I'd say it depends - for a data science type role then the maths/stats knowledge is equally important. For some applications of deep learning it's perhaps less so, there's also just less maths needed to get a general understanding of what is going on there and for people building products there isn't necessarily much need. As far as diffusion models and LLMs are concerned when the base models are so expensive to train then outside of the big labs the work is more building stuff around those models. You can fine-tune, you can add a vector database but a lot of the work there is building a web app or mobile app. There are some very successful indy devs in this space who certainly aren't maths guys, they are however experienced at building consumer apps and have spun up successful products using their web dev skillset.
And even within the few big labs making the base models (whether independent or whether within Google, Meta, X etc.) there are still other roles that don't require as much maths/ML knowledge - someone has to monitor/keep the servers running, there are roles for regular "software engineers" not just "research engineers" etc.
