My First Research Experience
i got a summer research position! it's at my own university, unpaid, but it's research.
i'm working with a professor on something involving natural language processing and honestly i'm still figuring out what we're actually doing.
the project
something about using language models to extract information from scientific papers. the details are still fuzzy to me but the general idea is: can we automate literature reviews?
my job: implement a baseline, run experiments, break things, learn.
week one revelations
research is not like coursework
in courses, there's a right answer. you do the assignment, you get it right or wrong.
in research, nobody knows the right answer. that's literally the point. we're trying to figure it out.
this is both terrifying and exciting.
reading papers is my job now
i spent three days just reading existing work. no coding. just reading, highlighting, making notes.
turns out you can't solve a problem if you don't understand how others have tried to solve it.
my mentor is a real person
professors are intimidating from a distance. but my mentor (a postdoc, technically) is just... a person. she gets coffee, makes jokes, admits when she doesn't know things.
somehow this was surprising.
the actual work
so far i've:
- set up the development environment (took two days, everything is painful)
- reproduced an existing paper's results (harder than it sounds)
- started implementing our modifications
- broken things repeatedly
the ratio of "coding" to "figuring out why code doesn't work" is about 1:4.
what i'm learning
1. reproducibility is hard academic code is often poorly documented. hyperparameters are buried in appendices. setting up the same environment is a whole thing.
2. small differences matter changing one hyperparameter or preprocessing step can change results significantly. science is precise.
3. negative results are still results "we tried X and it didn't work" is valuable information. failure is data.
4. writing matters research isn't just doing things. it's communicating what you did clearly. writing is part of the job.
the vibe
it's quiet. there's a small office with three desks. we have a whiteboard covered in diagrams i half-understand.
every day i feel slightly less lost. that's the metric of success right now.
is research for me?
i don't know yet. but i like the exploration. i like that we're trying to figure out something nobody has figured out yet.
ask me again at the end of summer.
ran my first successful training run today. loss went down. i felt like a scientist.