hi thots #5
7:38 pm, wednesday february 21, 2024

hello all! today, february 21st, i am attending aaai as an undergraduate consortium scholar here in vancouver, british columbia. one year ago, i attended aaai as a paper author for a workshop on ai and diplomacy in washington dc. in the nature of respecting these rare parallel moments, i thought i would pay an honor and give a quick research and academia summary and thoughts i've had from this past year.

since january of last year, i have been working on a project related to nlp and speech processing, and that's still continuing on. after the president botrick paper, it seems things are going a bit slow in the research alley on my part. a recent diagnosis of some mental illnesses may have contributed to that, a general stint of burnout and just general exhaustion from working so hard towards the end of high school and during my first year is probably another large factor. in fact, i feel almost regretful, now that my research at mila is coming to a close, that i didn't utilize all the resources i had. but i'll probably be back at mila at some points anyways, everyone in montreal in ai winds up there, and for good reason. it is such a fantastic place, it's just intimidating to be in such a researchy environment. that's how aaai and other big conferences can also feel sometimes, so it's always exhausting towards the end of the event.

let's talk a bit about how my research interests have changed. since chatgpt became a thing, i've been feeling awfully bored by a lot of what is happening in the nlp scene. it seems that the only research advancements are just the same models, but with more parameters (of course, this isn't actually the case, but it's what everyone seems to be talking about these days). for this reason, i've been leaning more into exploring other disciplines in ai. for a good amount of time, i spent a lot of my energy learning about music and ai. in fact, i had written a proposal on neural audio effects, deep learning based style transfers, and natural language to audio editing pipelines. of course, i have no idea how this research would be conducted if it were to be executed. but at least my heart was in it. i considered a career change at one point, even talking to a professor in illinois about some music and ai related things. lately, however, it seems i've returned to my interests in nlp. this is for certain because of a course i'm taking this semester called natural language understanding with deep learning. every assignment is picking apart neural networks and the various components of the model experimentation procedure in different ways, so it's a really interesting set of problems. as a result, i think i've been more exposed to more creative and intuitive advancements in nlp, different ways of looking at the same thing and trying to learn about what the next step is. and people have been really creative. in fact, this has led to a newer, really interesting sub-discipline of ai called "explainable ai", which is related to focusing on an analysis of the inner workings and thought paths and patterns that the model has rather than what its output is.

i'm going to go enjoy the rest of this conference. maybe we'll do this again next year :)