This is where I work out things I don't understand yet.
By day I build PageOn and iArt — tools that turn an idea into a presentation, or a sentence into motion graphics. Watching more than a hundred thousand people use them to explain things is an unusual thing to get to see, and most of what I understand about explanation comes from it.
Chiefly this: explanation has more than one axis. The tools I build work on what a person can produce, and how well — and they keep getting better at it. This site is about a different axis: that a finished explanation, however good, still has no way of knowing where a particular reader needs more. Nobody's does yet.
Three threads run through this site.
The wet one. My PhD was on reconstructing neuron morphology with machine learning, which meant several years of looking closely at neurons without ever really understanding what they do. I am fixing that — going back through the dynamics from vector fields up, slowly, and writing down the parts that took me longest. Most of what is here is that.
The dry one. Whether any of it transfers. The efficiency gap between a brain and a datacentre is real and largely unexplained, and there is a small body of work — oscillatory state-space models, spiking networks, neuromorphic hardware — that takes it seriously rather than rhetorically. I want to know which parts are substance.
The form. No author knows where a given reader needs depth. Written explanation has to guess, and guesses badly in both directions on the same paragraph — too basic for one reader, too fast for the next. So the pieces here expand in place, and you can ask your own questions inside them. That is an experiment, and it is the third reason this site exists.
I am wrong regularly, and the corrections have been the best part so far. If you catch something, say so.