Jev introduces a new shape of LLM - System One, aka Decision Models
Last week TypeSafe AI unveiled Jev , their first example of a new category of model that they are calling "System One models" (I'm with Maggie Appleton, I think "decision models" is a better name for these). Jev is an interesting variant on the usual LLM format: it still accepts text inputs, but instead of text output it returns floating point numbers corresponding to categories, yes/no…
TypeSafe AI has introduced Jev, a novel category of model they are dubbing "System One models" or "decision models". Unlike traditional language models that return text, Jev delivers floating point numbers representing categories, yes/no questions, ratings, and confidence scores. This innovation allows for more targeted querying, akin to unstructured function calls that yield probabilistic decisions.
Notably, Jev operates on a cost-effective model where input costs are charged while output is free, making it cheaper than even OpenAI's GPT-5 Nano. The system accepts various data formats as "state" objects, asks questions, and returns parallel evaluations of the queries, delivering results in a similar time to a single query. Jev’s utility shines in classification tasks like spam detection or prioritization, and it shows potential for search reranking by comparing candidate relevance scores.
However, its lack of explanatory output raises concerns about transparency and potential bias. The rapid community response indicates a significant level of interest and experimentation with Jev's capabilities, despite its recent release.
Written by urgent.news from Simon Willison's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
