Introducing System One Models and Jev
- Source
- typesafe.ai
- Author
- albelfio
- Date
Context
TypeSafe AI's Jev produces decisions in a structure software defines in advance. Instead of generating a written answer, it returns permitted values with probabilities, which surrounding code can use to classify information, route work or choose a branch. The company says it generates these outputs in parallel. Its introduction describes this as a model designed for decisions inside software.
The guarantee concerns output structure. A value can fit the required format and still be the wrong judgment. That distinction matters when interpreting the company's claim of eliminating : its stated zero-error figure refers to guaranteed schema matching, rather than an empirical demonstration that every decision is correct.
- hallucination — When a model states something false with full confidence — inventing facts, citations, or APIs that don't exist.
- calibration — How well a model's confidence matches reality — a calibrated model saying "90% sure" is right about 90% of the time.
- LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.
- tool use — A model's ability to call external functions — run code, search the web, edit files — instead of only generating text.
postTypeSafe AI launches Jev, a model for structured decisionstypesafeai- articleJev introduces a new shape of LLM - System One, aka Decision Modelssimonwillison.net
article[AINews] Jev: a “System One Model” that only decides/classifies/routes/scores — >100x faster, >200x cheaper than small frontier LLMswww.latent.space
Checking sign-in…
Loading comments…