Deciding on a path

Watched Welch Labs’ “The Dark Matter of AI” — much more concrete than Wednesday’s reading, actually showed what mech interp work looks like rather than just arguing that it matters. Went and found Neel Nanda’s quickstart guide and the Alignment Forum post on “How to Become a Mechanistic Interpretability Researcher” off the back of that. Both pushed the same thing: stop reading and start doing something hands-on, even something small, instead of trying to front-load months of theory first.

Also watched the ARENA Week 1 Day 2 intro-to-MI lecture on YouTube, mostly to see what a structured curriculum actually covers. That settled it — going with a curriculum-style approach (learnmechinterp.com, TransformerLens) instead of freeform reading. Starting properly this weekend.