1. Read software libraries first. If new techniques make it into standard implementations of major libraries, it’s a solid signal the new technique is actually

1. Read software libraries first. If new techniques make it into standard implementations of major libraries, it’s a solid signal the new technique is actually worth your time. 2. Don’t bother with arxiv / conferences / etc., until a paper is at least a year old and remains highly cited & investigated. In DL especially, papers are often slapdash permutations of existing ideas applied with one weird extra trick that ekes out “SOTA” results in one benchmark. Architectural choices are often not motivated by any evidence, and whatever scramble of grad student descent was needed to arrive at the final architecture choice is conveniently left out of the paper (and often would call the result into question from a multiplicity of testing POV). 3. Read what you enjoy. It shouldn’t feel hard to stay current in your main area. If it is feeling hard, you’re trying to force it. Maybe pick a sub-specialty you enjoy and only focus on that for a while. 4. Forgive yourself for not staying current. You are busy, reading research papers takes time & is not often the best use of time compared with competing interests. This is normal. If some situation / job interview / academic program is making you feel otherwise, it’s probably a red flag about that thing.