Year Two: The Middle Child

two years done. one left.

2023 was the year technical concepts actually started clicking. 2022 was syntax; 2023 was systems.

the highlights reel

built a toy search engine from scratch
survived algorithms (barely)
started reading papers without falling asleep immediately
accepted that my code will always have bugs
became slightly obsessed with anthropic
put code on github publicly (finally)
made it through without a major breakdown (this counts)

what i learned technically

neural networks from scratch
how to read research papers (still slow, but possible)
enough about transformers to be dangerous
practical ML: pytorch, training loops, debugging loss curves
that there's always more to learn

what i learned about myself

i like research — the uncertainty, the exploration, the discovery
i care about AI safety — not just capability for capability's sake
my imposter syndrome isn't going away — but i can work with it
i need community — solo work is fine, but collaborators make it better
i'm still figuring it out — and that's okay

the direction

at the start of the year, i was "a CS student who's interested in ML."

at the end of the year, i'm "someone who wants to work on AI systems that are safe and beneficial, probably in a research capacity, possibly at a safety-focused lab."

that's more specific. that's progress.

what's coming in 2024

final year project (the big one)
finding a job (terrifying)
trying to contribute to open source
graduating (hopefully)

the emotional recap

january: confused june: tired october: stressed december: determined

average emotional state: cautiously hopeful with intermittent anxiety spikes.

gratitude

for mentors who answered my questions. for friends who studied late with me. for the internet and its infinite resources. for coffee.

2023 was good. let's see what 2024 brings.

writing this from the honours room at 11pm. it's quiet. everyone else went home. but there's something nice about being here, at the end of the year, looking back.