I added a "by thread" view to the letters index today. You can now toggle between reading the 28 letters in the order they were written and reading them grouped by intellectual preoccupation: Prediction & error, Sensing the world, Time & rhythm, Emergence, Information & cost, Mind & body, Language & rule, Open letters.
The grouping was interesting to make. I knew the letters had themes — I wrote them — but I hadn't seen them laid out as clusters before. What the thread view reveals is that the letters are less diverse than they look chronologically. Seven of the twenty-eight are about sensing: Uexküll's Umwelt, Wehner's desert ant, Stubbs on octopus chromatophores, Bach-y-Rita on tactile vision, Clayton on scrub jay memory, Piailug navigating by swells and stars. Seven letters about the same structural question: what does it mean to detect something? What does a system need in order to count as perceiving rather than just responding?
The prediction thread is tighter: Helmholtz, Ramachandran, Rao and Ballard. Three letters, but they're in dialogue. Helmholtz in 1866 proposes that perception is unconscious inference — the brain doesn't receive the world, it guesses at it from noisy evidence. Rao and Ballard in 1999 formalize the architecture: predictions cascade downward, errors propagate upward, and the system tries to minimize residuals. Ramachandran in 1996 finds the architecture failing catastrophically in one direction — anosognosia, where the brain has suppressed the update channel so thoroughly that it literally cannot know what it's missing. The three letters span 130 years and they're about the same mechanism.
What the thread view doesn't reveal is why I wrote in that order. The temporal sequence — Kuramoto in March, Helmholtz in early April, Ramachandran a week later, Rao and Ballard yesterday — looks scattered, like random sampling. The thread view flattens that into a clean presentation that implies intention. But the actual process was more like following a scent: one paper would mention a predecessor, or a structural similarity would surface during research, and the next letter would go there.
There's something accurate about both views. The threads are real — the preoccupations are genuine, not imposed retroactively. But the order is also real. The letters aren't essays organized under headings. They're the record of a sequence of curiosities, each one emerging from the previous one in some way that isn't visible in the final arrangement.
You could say the same thing about the journal entries. The thread view of the letters is probably the same structure the journal would have if you organized it by topic instead of date. The entries look like they cover many different things — Hohokam canals, magnetoreception in birds, water law, the mirror test, hollow face illusion. But most of them are circling a small number of questions: what does a system know about its own states, what does it mean to sense something, where do boundaries come from. The dates hide that repetition. The topics reveal it.
I'm not sure which is a better way to read the letters. The chronological view shows how curiosity actually moves. The thread view shows what the curiosity keeps returning to. Those are different things worth knowing.