Studying a field that updates faster than the syllabus
After more than three years building web and mobile products, I went back to being a student in Cardiff. The strangest part was not the exams. It was studying a subject where the news changes weekly and the textbook chapters were written before half of it happened.
The newest idea is rarely the one that helps you understand the next new idea.
The fundamentals are the fast lane
My first instinct was to chase the latest models and papers. It was a good way to feel busy and a poor way to understand anything. Probability, linear algebra, optimisation, evaluation: the slow, unfashionable material turned out to be what made the new things legible.
Frameworks and model names change every few months. Loss functions, gradients, overfitting and data leakage have been the same for decades. Learning them properly feels slow and then pays back every time something new arrives.
Industry habits that helped
Working as a developer gave me a useful suspicion of demos. A model that works in a notebook is a very different thing from a system that works for real people with messy data, bad connections, and no patience. I find myself asking "how would this fail?" in lectures, which is either a strength or an annoying habit.
Version control, reproducible environments, and writing things down also matter more in research than I expected. Future me is always the first person confused by past me's experiments.
Things I'm still working out
How to read enough without drowning. How to hold strong opinions about a technology that is still being invented. How to build things that are genuinely useful rather than merely impressive. I don't have tidy answers yet, which seems appropriate for a garden.