As a grad student (ML) close to finishing I can say exactly what I want. I want to read math books, program, and publish slowly (papers + blogs) with meaningful and substantial work. I do not want to be chasing benchmarks or worrying about how to play the politics of paper publishing to get through the provably random[0][1][2] noise that is the review process. But no one is going to pay me for this, including academia. I know a lot of my peers are chasing big paychecks but I don't understand why anyone would go through all this just for money. There's much easier ways to make money. If you know of anyone that would actually hire me for this, please do let me know.
That said, I don't regret doing my PhD. There is a lot of personal value in being able to (mostly) freely study a topic in extreme detail. Obviously you need some obsessive behavior to do this. The thing is that I just want to keep doing it. But with a focus on the learning and extending human knowledge part and cutting out the bullshit.
[2] Personal experience: I've had reviewers state that a paper with >100 citations is not useful to anyone while many of those citations are from hard sciences using it for explicitly the reason we made it. In another review round a reviewer asked us to include experiments comparing to our main comparitor, which was included in every single graph and table we had in the entire paper (no ability to respond). I've seen rampant abuse of the review system (ACs accepting and rejecting papers in weird ways), collusion rings, and overall benchmark chasing (which hinders a lot of research all together). We are encouraging lazy reviews, everyone agrees, but few are actively trying to make a difference. The worst part of this is that the people hurt the most by these actions are in fact the grad students. Their graduations depend on top tier publications and their ability to get their first jobs highly depend on their top tier publications. Ironically being hired by people who actively discuss issues with the publication system and how much noise there is. It is frustrating.
Government labs, scientific agencies (NASA et al), or any FFRDC might be a good fit. Also old-hat industrial labs. Defense and defense contractors can also be good, if you don't mind that type of work. They don't care much about vanity metrics like publications at top X conferences, and they don't move at the fast pace of product-oriented groups in tech companies.
You can also start an LLC and fund it through SBIR/STTR and/or transition-to-practice style grants, but you'll have to hire a professional admin staff and pay them more than you're making if you want the type of life you're describing (and it'll take some years).
Gov labs aren't too bad, but funding often comes with a lot of conditions. Generally you have unconditional funding for weapons but if you can find your own funding you can research whatever you want. I'd rather not do weapons and it is a reason I've moved away from that route. Also these are mostly hard science type research (physics, chem, climate, etc) and less interested in doing things like building general intelligence. SBIRS are typically too high of a TRL but STTRs are a good route and tend to be more basic research. I am looking into that. But I've also been in the SBIR/STTR life and boy is it a lot of writing.
That said, I don't regret doing my PhD. There is a lot of personal value in being able to (mostly) freely study a topic in extreme detail. Obviously you need some obsessive behavior to do this. The thing is that I just want to keep doing it. But with a focus on the learning and extending human knowledge part and cutting out the bullshit.
[0] https://blog.mrtz.org/2014/12/15/the-nips-experiment.html
[1] https://inverseprobability.com/talks/notes/the-neurips-exper...
[2] Personal experience: I've had reviewers state that a paper with >100 citations is not useful to anyone while many of those citations are from hard sciences using it for explicitly the reason we made it. In another review round a reviewer asked us to include experiments comparing to our main comparitor, which was included in every single graph and table we had in the entire paper (no ability to respond). I've seen rampant abuse of the review system (ACs accepting and rejecting papers in weird ways), collusion rings, and overall benchmark chasing (which hinders a lot of research all together). We are encouraging lazy reviews, everyone agrees, but few are actively trying to make a difference. The worst part of this is that the people hurt the most by these actions are in fact the grad students. Their graduations depend on top tier publications and their ability to get their first jobs highly depend on their top tier publications. Ironically being hired by people who actively discuss issues with the publication system and how much noise there is. It is frustrating.