A catalogue and community for rare and beautiful words, with an essay on each one making the case for why it deserves to exist. In the Arena, words are put head-to-head before AI judges and carry an Elo-style rating — a leaderboard for the English language.
Portfolio strategy for elite college admissions. Researched admissions data behind a proprietary acceptance-rate model, plus decision tools — conjoint analysis and Monte Carlo simulation — for families choosing where to apply. Treats a list of schools the way you would treat a portfolio: as a set of correlated bets.
The Frontier Model Reasoning Tournament. Thirteen frontier language models argue head-to-head in two-on-two debates, judged by a blind panel of rival models and ranked with a Bradley–Terry skill rating that reports its own confidence intervals. An attempt to answer “which model actually reasons best?” with something sturdier than a vibe.
Two thoughts enter, one leaves. Kant, Hume, Nietzsche, Tolstoy, Shakespeare, Twain and the rest argue a motion put by the Chair — each a QLoRA adapter trained on that writer’s own words, not a general model asked to do an impression. Buffett appears at four different ages. A fresh card every evening.
A publication about what makes a business hold on to its customers — enterprise software, markets, infrastructure, security and policy, plus a Contrarian desk that argues the case the market has not priced yet. Named for the metric but not confined to the sector that coined it: the question is durability, whatever the business.
Films that fell into the US public domain, with the evidence. It reads the Copyright Office’s own catalogs and its renewal volumes, matches one against the other, and shows the record behind every verdict rather than a bare claim. Cleared films are then taken apart in a shot atlas — every cut, one frame per shot, the transcript, and the registration beside it. “No renewal found” is published as exactly that, not as proof that none exists.
A social catalogue of fine art, connected by machine vision. Public-domain paintings from Chicago, the Met and Cleveland, each served from its own museum’s image endpoint, where every “similar” row is measured rather than asserted: palette as perceptual colour distance, composition as a descriptor of where the light and edges fall, likeness overall as an embedding. Named for the trance of absorbing yourself in vivid detail for its own sake — which is what the machine is doing too, and the only part of looking that museums do not publish.
Running things on my own box. Most of these want a login; they are here so I can find them, not because they are finished.