A Garden, Not a Blog
Why this site publishes unfinished notes on purpose — the three tracks, the status system, and the ADHD-proof pipeline behind it.
This site is not a blog; it’s a digital garden. The difference comes down to one thing: a blog’s unit is the finished article, a garden’s unit is the growing note. Every post here wears a maturity badge — 🌱 seed for a rough, freshly planted note, 🌿 growing for an understanding that’s taking shape, 🌲 evergreen for something stable enough to cite. Seeds never have to apologize for being rough.
Why it works this way
I have ADHD. For me, the task “write a complete blog post” has an activation cost so high it almost never happens — but “prune today’s paper-reading notes and put them out” is doable. So the core rule of this garden is: I never write from scratch. Every post is a byproduct of study — paper deep-dives, course notes, experiment logs — planted in its roughest viable form, then pruned a little on each revisit.
Maintenance follows the same principle: publishing is compressed into a single git push, and everything after that (build, deploy, going live) happens automatically. I tend this pipeline together with my agent (Claude) — it turns notes into drafts, checks facts, and handles the release chores, while I do the actual understanding and judgment. Which makes this site a small living experiment for the Loops track, too.
The three tracks
- Latents — generative models: diffusion, image & video generation, conditioning, efficiency.
- Logs — systems: MIT 6.824 distributed systems, Designing Machine Learning Systems.
- Loops — agentic systems: tool use, orchestration, evaluation.
Why these three? Are they actually related?
They are — and the intersections are the point. Generative models are becoming infrastructure, so questions like “how do you serve a diffusion model cheaply” or “how do you give an agent a reliable image tool” are Latents, Logs, and Loops questions at the same time. My bet: the people who understand both the mechanisms and the systems will be standing in the most interesting spot.
How to read this garden
Browse by track or by series, not by date. Timelines mean little in a garden — a seed planted in June might turn evergreen in December. If you spot an error, find me on GitHub; pointing out mistakes is the kindest thing you can do for a gardener.