$ man jev
// what is jev
Jev is TypeSafe AI's
"System One" model. It doesn't generate text — you give it a state (some text or JSON)
plus a set of typed questions (yes/no, multiple choice, or a score), and it returns
calibrated answers in roughly 100–300ms.
// why it's interesting:
- → 20–200× faster than a normal LLM call, since it's not writing prose
- → fractions of a cent per call — cheap enough to run on hundreds of items
- → answers are typed and probability-calibrated, so software can act on them directly, no parsing
// source for everything below: github.com/raaulc/jev-projects
$ ls -la ./jev/
// daily-automated arxiv paper ranker — jev scores every new cs.ai/cs.cl/cs.lg/cs.ma paper for relevance and novelty, github actions cron keeps it fresh every morning, full date-wise history browsable
// status: in progress
// five famous ethical dilemmas run through jev's typed-decision api — real/not-a-dilemma, ethical framework, difficulty score, and an actual yes/no verdict for each
// status: complete