@ryanvogel built what they call an LLM from first principles with Jev: 29 yes/no questions per character (a–z, space, comma, or period). The highest probability is appended and the updated text is fed back, an autoregressive loop made of a classifier.

vogel (@ryanvogel) published this on 2026-09-16. @ryanvogel built what they call an LLM from first principles with Jev: 29 yes/no questions per character (a–z, space, comma, or period). The highest probability is appended and the updated text is fed back, an autoregressive loop made of a classifier. i made an llm from first principles with Jev 29 yes/no questions per character: should the next key be a–z, space, comma, or period? How Jev or TypeSafe is used, in the post's own words: “i made an llm from first principles with Jev 29 yes/no questions per character: should the next key be a–z, space, comma, or period?” The post does not name a separate audience. It is for people reading what @ryanvogel shows here (it quotes @CompleteSkeptic; it includes media).