Something I’ve realized recently is just how time consuming research can be. The whole process. Every project relies on how hard and how quickly you cycle iterations, come up with new ideas. And among other things, that requires constant vigilance. If I could express my frustration with the process generally, and I’m not really sure it is frustration, it would have to be with myself, because Daniel and every CS professor I’ve met here has only been kind and supportive, and makes me want to drop everything else and take their classes and pursue their subdisciplines. But I think most importantly, by talking to older grads, my friends, and others who have obtained a certain level of research maturity, I self-diagnose my thinking as commitment issues. The main symptom would have to be the erroneous act of approaching any research itself with the lens of “oh yeah, I’ll make sure the project is interesting to me, the project is relevant to ongoing SOTA, the people that did this type of research in the past went on to do x and made y, and so on, treating it like a list to be ticked.
I do think that at the bare minimum, having a question that is genuinely interesting to you is super important. That, and being fortunate enough to have an awesome advisor. But I think as I find myself increasingly entrenched in the process, I’ve realized that there are a whole lot of other factors I may not have been considering that makes any given question worth pursuing, but that in my haste, my thinking was basically an excuse to not get committed fully to a project. More specifically, everyone that I’ve talked to has consistently said to just dig deep into something that seems interesting to you, and not to worry about the citations on a paper, H-index or whatever. Regrettably I’m not yet fully convinced, but in time maybe will be.
One of the reasons I look forward to grad school is that I can just focus on my questions specifically. That’s not to say I haven’t really liked my classes, but these days, I always feel like I want to try to escape being a student, because there seem to be so many opportunities out there, and yet the time needed to be spent on lectures, projects, labs, etc. is nontrivial for sure. And the thing is, sometimes I wish I had like perfect knowledge of DL, graphics, systems, optimization so I could just go and do research – one of my friends has extensive knowledge on algos and nlp and literally only takes intro art and writing classes so that he can put all his time into language understanding research and AI safety. To get more time, I’ve tried to cut out as many small, insignificant things as possible, the odd email or application that feels great to check off the task list because it’s easy to do but not really that important, essentially a dose of dopamine. But to me what’s really difficult is getting to a mental space where I can really attack a subproblem in the pipeline repeatedly, like hitting yourself over and over again, for the short-lived moment when you actually get decent metrics, and that feels great. But since you never know when that will come, along the way you’re just navigating a bunch of local minima, except the difference is that you’ll never quite know exactly which direction to move in.
. . .
There’s a particular room on the seventh floor of the Barus and Holley building. The room faces the south, and on clear days, you can see a mile or two out into Fox Point and an ocean-like expanse that is actually just the Providence River. Recently, it’s been snowing like crazy, and there was incredibly nice-looking sunset today during my lock in session:
. . .
Actually, maybe it’s easier for me to write like this. Basically pieces that I jot down occasionally, shower thoughts here and there. The structure of Vonnegut’s prose, but worse content fs. Sometimes people talk about the notion of “high literature”, books that have artistic merit, you know will open your third eye, give you superpowers, all that stuff. I don’t really believe in it, but what I do believe in is like the direct opposite, which would really be the book equivalent of online brainrot slop. I had the misfortune of reading one of these recently…