What people are shipping with Jev.

jev playing connect four completely pointless but i am still into it @typesafeai https://t.co/vKsFMmnSCQ https://t.co/8zd5Va4oLK

instead of generating text, jev from @typesafeai generates structured output this makes it great for classification tasks like model routing, tool selection/search, and guardrails

Got early access to @typesafeai "Jev" and my first thoughts went to SQL (and @duckdb). SQL is the one query language everyone already knows and DuckDB is my favorite data wranglin

@CryptoPraetoria @ownerofmarket @AsideAI @typesafeai Jev is better for browser navigation and execution. It chooses only from real page elements with typed decisions in ~70-500ms,

Time taken seems to be increasing over time, @typesafeai ' JEV seem to be getting busier 👀 Try https://t.co/V7CgA6iISc https://t.co/IbZapgMPVx

@typesafeai @parsewise 2. Jev was a lot more literal when following instructions, meaning I had to tune the task guidance more so than LLMs. E.g. explaining that small indirect c

Jev for first pass, LLM for the gray zone @typesafeai ftw https://t.co/tkVoCWYQ2U

@typesafeai Finding 1: Jev improved overall reliability, but not uniformly. Without Jev: 20/50 passes With Jev: 24/50 passes 4 faults improved, 2 regressed, and 4 unchanged. This i

AkihikoWatanabe/paper_notes · issue · open · 3 comments | https://typesafe.ai/blog/introducing-system-one-models-and-jev

@rauchg @typesafeai @vercel Default-auto with a safety reviewer on every command is the agent posture I want more CLIs to copy. @rauchg on @typesafeai fx (auto mode, reviewer on G

@o_kwasniewski @typesafeai is Jev classifying pass/fail after the run, or driving the clicks? a classifier I can slot in. a Jev-driven UI loop is a different harness.

@JoshARosen @typesafeai jev watching coding agents for progress/tests/drift is free senior qa energy. intervention precision matters - unnecessary intervene % vs missed drift

@langfuse @typesafeai @langfuse The certainty signal is what I'd push on. How does Jev behave when confidence is low but it still has to return a category? That boundary between "u

@hakimuddinkika @tamarajtran @typesafeai Jev is TypeSafe AI’s flagship System One model. It takes program state plus typed questions (choice, score, or noul/yes-no) and returns str

@MrAdetilewa short version: jev doesn’t write text. you give it options and it picks one, with a confidence score made a silly magic 8 ball so you can feel that in like 10 seconds

@typesafeai Finding 3: Jev can only rank the hypotheses it receives. When Luna proposed the wrong set of explanations, Jev could approve a coherent but incomplete story. Every requ

@typesafeai Jev is moving fast - browser use, context compaction, RAG, agent harnesses, even a programming language. I went down the rabbit hole on how it works and why these use

@fazxes @typesafeai vercel fx safety mode on jev is interesting. false allow rate at the prod threshold beats another relative speedup chart

👀 55 people right now exploring JEV use cases at https://t.co/z0izy4XNao @typesafeai https://t.co/41QGinihaF

@hazemomier @typesafeai This is just an experiment to showcase what Jev is capable of. A production use case would need more research, and real traffic to test against, before I co