TypeSafe AI's Jev Skips Text Generation for Typed Decisions

TypeSafe AI released Jev today, its first System One Model, in early access. The claim: intelligence comparable to existing LLMs on System One tasks, two orders of magnitude faster and more efficient, no string generation — “unstructured state in, typed probabilistic decisions out” — and no hallucination.

That trade is the interesting part. Chat models have been good for years and almost nothing runs itself, because fluent text is not something software can call. Calibrated probabilities are.

Claims like that need measuring, not reading; two orders of magnitude stays a launch-day number until someone reproduces it. My test: wire it to the one nagging alert in my house and compare its probabilities against the calls I make by hand. The pitch is a function call for decisions, and function calls are testable.