Spent the past few weeks building out the Notes — 15 entries now, covering everything from Markov chains to flow matching.
The exercise of writing these forced a distinction I hadn’t articulated before: there’s a difference between understanding something and having a story about it. A formula you can derive isn’t the same as a formula you can explain to someone walking next to you. These notes are my attempt at the latter.
Some connections surprised me. The policy gradient note ended up referencing martingales — the baseline trick in REINFORCE is really just a martingale difference argument. I wouldn’t have seen that if I hadn’t written both notes in the same week.
The format is deliberately terse: no textbook preambles, no “in this section we will discuss.” Just the idea, why it matters, and where it connects.