CommunityJev Media & Talks
Jev outputs probabilities, not text
Telusko Alien Coach on why generating text is overhead for a classifier, with Vercel's reported speedup and Armin Ronacher's objection.
- Format
- Video
- Published by
- Community
- Author
- Telusko Alien Coach
- Last verified
- Sep 2026
Used inJev for Software & DevTools
Highlights
- The description dates the release to 18 September 2026 and calls Jev a transformer that outputs probabilities.
- It says Vercel reported a safety-classifier swap that was five to eighteen times faster on their workload.
- It quotes Armin Ronacher, CTO of Earendil, that the design delegates the hallucination problem to the user.
Watch out
The Vercel range and the Ronacher line are the video description. The public launch posts are dated earlier than 18 September.
Questions about this entry
Short answers drawn from the linked project or page.
That range is the Vercel figure as repeated in the video description, for one classifier swap, and the entry flags it as unverified restatement. Latency multiples depend entirely on how much state you send and how many options you score.
That the model can only return one of the keys you defined, so it cannot invent an answer — but it can still hand you the wrong key with high confidence. Where you draw the accept threshold, and what happens on the tail, stays your problem.
Do not rely on it. The description pins the launch to a specific day, while the entry notes that the public launch posts are dated earlier. Treat the description's date as approximate and go by TypeSafe's own announcement.
The source does not document it. The capture is the title plus YouTube's footer, and the entry's own copy covers only the probability-output framing and the speed claim. Nothing here says whether non-text state is accepted or rejected.
Telusko Alien Coach published it, describing Jev as a transformer that outputs probabilities rather than text.
Armin Ronacher, CTO of Earendil, quoted as saying the design delegates the hallucination problem to the user. The entry records that line as the video description's.
Because for a classifier, producing text is unnecessary overhead. The description frames Jev as outputting probabilities for a fixed question instead.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
YouTube: Telusko cites Vercel's 5–18× safety-classifier swap and Armin Ronacher's objection.
YouTube: LangChain maps System 1 and System 2 onto fast decisions versus slow generation.
Facebook: Intake Router keeps diagnosis out of the answer space at the front desk.
FacebookClinicnotes
YouTube: Jev explained for routing and judging, with a comparison against LLMs.
More Jev resources like this
Riley Brown on a decision model that cannot write: a model router, Choice, Score, and Noul, then speed, cost, and where to get access.
Jev Media & TalksVideo#community#video#explainer
Community
LangChain maps Kahneman's split onto models: System 1 as the fast automatic decision, System 2 as the slow deliberate one, with Jev in the first role.
Jev Media & TalksVideo#community#video#explainer
Community
Tim Lin's visual explainer on System 1 versus System 2, the Choice, Score, and Noul primitives, and how JevScout uses Jev inside a coding agent.
Jev Media & TalksVideo#community#video#explainer
Community
Joe Maddalone on using Jev for routing and filtering before a generative model, with demo code in a transcript-analyzer repo.
Jev Media & TalksVideo#community#video#routing
Community
TJ Robertson on the jobs that fit a classifier with a confidence score, and the speed and cost he trades against a frontier chat model.
Jev Media & TalksVideo#community#video#routing
Community
Max Jesch, on IBM Developer, walks Jev as a System One model that returns probabilities for classification instead of generating text.
Jev Media & TalksVideo#community#video#explainer
Community
From the Jev community
Posts from builders shipping with Jev right now.