this is also applicable for those working in industry! some other things i've found to be helpful as a practitioner to keep up with ML literature:
* pick subfields intentionally -- doesn't have to be relevant to work
* focus on themes in papers, not necessarily specific papers

> "pick subfields intentionally"
i intentionally follow things relevant to my work (continual learning) and other unrelated things that bring me joy (breaking neural nets, privacy, music)

> "focus on themes in papers"
i think this is where research & practice differs most. in practice i have a broad problem (ex: robustness to distribution shift) and need solutions. if an idea is developed in multiple papers, i have more confidence it will work for my use case

as a practitioner, rarely do i need to be at the forefront of anything super algorithmic. i want simple solutions that i can implement quickly without large changes to the codebase. but i wish i knew of and had opinions on the latest & greatest algos like other cool researchers.