free judgment for all
controlled by nobody.
every AI model has been trained on what people wrote. none has been trained on what people judged, with the reasoning and the outcome attached, because that record has never existed at the scale of a market. the people who judge for a living, in markets, courts, labs and legislatures, keep the verdict in one place, the reasoning in another, and the score nowhere.
we are building that record. anyone can strike a judgment into it across six facets: who is judging (identity), what they took in (input), how they reasoned (method), what they concluded (output), what follows at scale if they are right (scaling), and who may read it and when (transmit), with a timestamp for when it was struck and one for when it was transmitted. any later entry that links to an earlier one can score it, by a method it states. no owner and no official scorer. one open record, valued by use.
how it works
after compute, the scarce input to intelligence is scored judgment. a market price is the oldest form of it, a verdict scored by the next verdict. the beier tensor subsumes it: it keeps the price and adds the reasoning, the part an AI model can learn from. a price that carries its own judgment is a free market price. people strike judgments into the record; AI models train on scored judgment; i expect better AI models to make the record worth more to strike into. cognition and compute feed each other, and the loop runs on a substrate nobody controls. that is how we fold space. information that once took an institution a generation to accumulate can arrive in one record, scored, open to all.
if this works, each struck judgment can improve the ones that come after it, for people and for AI models alike. i expect the cumulative effect, as more prices are struck as free market prices, to be a step change in how well we coordinate, above all in how we scale technology.
the bet, stated as a bet
the share of economic value that depends on people coordinating cannot be learned from scraped text. it can be learned from a scored record of judgment. that claim is deposited as a conjecture with the test that would falsify it. sixty-two papers build toward it. the method applies to me first.
the protocol is free.
i believe this is the best way to splice AI into our world: strand into strand, tensor by tensor, so the join holds.