When I was a Ph.D. student (mid 2000s), I was very in tune with my sub-sub-field, and would attend at least some of the key conferences and actively monitor the key journals for anything not already on my radar. Plus I was in a research group with 20 other people so things i missed easily came up in discussions.<p>Now I'm much more of a generalist, and because I work in ML instead of niche physics, there is a whole lot more to go through. So mostly I don't: I get a few newsletters that let me select new papers I'm interested in, and my work takes me down rabbit holes that result in deep-ish rabbit holes where I do the usual find some key papers based on google search and then go through the references (and citations) to get more papers in that area.<p>Otherwise, I find that the gap is very small between when something new comes out and when it's rolled into some of the common frameworks e.g
Pytorch lightning- although that's discipline specific obviously. Come to think of it, I also look at papers with code, the benchmarks section, to see what the state of the art for different datasets is. And if I have time I look through accepted papers at the key conferences.<p>Anyway, if i had to summarize (my experience only): researcher -> go to as many conferences and talk to as many people as you can; practitioner -> don't worry about missing this week's buzz, and dig in as required.
For quant finance, by looking at recent papers on SSRN, arxiv (there is a quant finance section), by getting email updates from finance journals of their new papers (often the working papers can be found online), and by subscribing to some free newsletters from New Economic Papers <a href="http://nep.repec.org/" rel="nofollow">http://nep.repec.org/</a> .<p>A caveat is that one can spend so much time keeping up-to-date that one's own research languishes.