Jev Experiment

A repository whose README and GitHub description both say only 'Typesafe.ai experiment'. Catalogued for discovery; the digest shows no code details, setup or stated purpose.

An experiment that drives Jev as an autoregressive next-token predictor: each step offers potential next tokens as choice options, and the resulting sequence of picks is the generated text. The author's verdict: it kinda works.

Text generation with a model that never generates text: jev-gpt asks Jev one typed question per choice — word type, WordNet category, a group of about 200 words split by embeddings, then the word — about 400 calls and 2 cents per prompt.

An experiment in two parts, done in one day: SATB part-writing exercises given to Jev with no musical hints and graded by rules, compared with Claude models on accuracy and cost, plus a piano mode where Jev picks chords forever.

Coaxes free text out of TypeSafe's classification endpoint by decoding one word at a time: every word is a separate choice question over a vocabulary, and the loop feeds its output back as the prefix. The README itself calls the approach a demonstration, not a good idea.

Novelty npm package whose crypto.randomUUID() look-alike sends 15 Noul questions plus one Choice to Jev and packs the returned probabilities into the bytes of a UUID v4-shaped string; explicitly not cryptographically random.