The first HN link in my comment addresses that. The short version: learn the earliest asms first, then progressively learn the newer ones until you get to today, and none of the knowledge will be wasted. Kind of like fast-forwarding.
The smallest, simplest, 'useful' (in terms of useful enough that lots of devs did good work with it and thus it might also be 'popular') ASM sets are probably also sufficient to start with. Provided you've got a good guide to using them, and also ideally a good sheet for why given instructions are packed the way they are in binary.
I do agree you're more likely to find pointers to such resources in more classic architectures. They're also more likely to be easy to find free copies of useful literature.
i.e. https://esolangs.org/wiki/FlipJump
Flip Jump is amazing. I understand the theory and how it works but it still amazes me that it does. Things like this is why I love the science in computer science.
And subleq even has a c-compiler and operating system, just wow. https://en.wikipedia.org/wiki/One-instruction_set_computer#S...