X posts

Curating clinical variables from free-text notes is tedious. General LLMs can help, but processing thousands of notes can be slow and costly. Inspired by Jev and the open-source community, we’re releasing MedJev to turn clinical notes into structured fields on consumer GPUs. A 0.8B model + 43 MB LoRA adapter, trained to extract 11 predefined clinical variables. On our benchmark of 2,895 held-out n

What if AI could art direct your website instantly? I gave Jev control of a design system and started throwing vibes at it: “Minimalist.” “Brutalist.” “Playful.” “Premium.” ⚡ Changes almost instantly. No generated React. No generated CSS. No waiting for an agent to rewrite code. Same site. New vibe. Jev just makes the design decisions.

i accidentally made an agent that maps out my entire app and renders every possible use cases of Jev https://t.co/gN57vV3Qja

Jevで、セリフの文章をリアルタイムに解析して、文節ごとにキャラの表情画を差し替えるシステム。自分が話すセリフに感情を込めるのと、相手が話すセリフへの反応と、両方やってます。 フレーム補完もつけて、バグも直し、前回公開版より、かなり自然になって来ました。ツンデレ、ドSモードにも対応。 https://t.co/vjD7mEq1Ni

Two AIs. One game. I let @typesafeai 's Jev and @brainFnCl's Laya play the same arena survival game where every enemy’s decision is made live by the models. How it works: the game turns each moment into one sentence and asks one typed question. Back come probabilities, not prose. About 30 decisions a second. Laya: 322M params, Apache-2.0, ~21ms per decision, $0 per call. Runs on my laptop. Jev: ho

Jev を使って、CNPトレカの種類を特定して、同時にそのトレカが存在していることの確認をするアプリを作ってみました。 単純にLLMに画像認識をさせるよりも圧倒的にはやくて、しかも光の映り込みにも強くて精度が高い気がする。 持ってるカードを使ってアプリ内のゲームで遊びたい時には、これくらいのものでいい気がする。

@typesafeai we are tested Jev as the decision maker for a Unitree G1 humanoid in Automata Room. The task: carry 4 crates to a storage rack, one at a time, then return to the entrance. Fewer, shorter trips score better. Full run below 👇 https://t.co/xpliAUN5fR

今話題のJevをNotionで試してみた🤖 一言で言うと、自動化と人の判断の境目を数字で引くツールなのかなと個人的には思った👀 受信箱にメモを1件書くと、確率で種別を判定して、基準を超えたものだけタスク・意思決定・ナレッジに自動で振り分けられる仕組みを作って、結果1件1秒くらい🔥 止めるべきものを自動処理したのは20件中0件 逆に念のため確認が3割 Jevに関してXで最近流れてくるけど正直あまりよく分かってなかったので、実際に試してみたら面白かったし改善の余地も見えたので引き続き試していこうと思います‼️ 検証詳細はリプに👇

I was exploring Jev by @typesafeai AI for safety use cases and it is an impressive and useful model, however don't expect it to magically auto classify any data without proper parameters and thresholds (see finding below). I open-sourced a scam-email classification where you can use your own API key to try it. What I built - a proof of concept for suspicious email classification aimed at scam emai

做了个小猫生存游戏 让官方 Jev 和本地 Laya (mlx)一起保护小猫 每一步由模型自己选 小猫会因此吃饱、受伤，或者饿肚子 结论：Jev 更准但慢，Laya 极快但不准 三种关卡各测三次 官方 Jev 通关 3/9 局 全部来自找饭关； 本地 Laya 通关 0/9 局 Laya 虽然每局都回过家，却没能满足各关的饱腹、健康或保暖要求 兄弟们说关卡设计是不是难了点？ 后面会做个 Jev-Like 决策模型闯关排行榜

I plugged @typesafeai's Jev into PaperDance and my arXiv feed stopped being wrong. Same 100 candidates. One yes/no question per paper. Off-topic cards on page 1: 4→0, 9→0, 7→0. ~1 s and $0.0004 a page.Left: before. Right: after. Real production data. https://t.co/lKVkMTBKzO

Jev is the FASTEST with huge amounts of data! So I decided to build Index, instantly searches over 100 posts per profile, identifies your niche and shows your rank among the top voices in the space. Just being able to achieve these results in a matter of seconds is next level. Really loved building this and you can try it yourself here: https://t.co/oZmtZM52Dw Feel free to tell me your thoughts ab

this JEV + Picsart split is f**king insane for AI workflows so i rebuilt one campaign around a simple rule: cheap decisions happen before expensive generation. [here’s what’s actually happening:] 1. JEV sits in front of the stack and decides `allow_subagent`, `ask_human`, `stop_retry` or `reuse_cache` before anything expensive runs. 2. Picsart handles the heavy work across 188 models from 34 provi

jemo 4: someone was like "jev is cool, I wonder what other old ideas are worth reevaluating?" so i guess i made a 1980s Expert System that runs on lisp + jev??? ¯\_(ツ)_/¯ https://t.co/wWFtPexxNi

“Where was that tweet I read yesterday?” I built Seen with Jev from @typesafeai to find it. Press ⇧⌘Space. Type what you remember. Open the saved screen. It works across Mac apps and displays. Jev ranks the results using your own TypeSafe or Vercel AI Gateway key. Local search works without a key. It is open source under the MIT license. https://t.co/DODuwvWAhL

Qwen3.5-0.8BをLoRAで追加学習して、選択肢を採点するJev風の判断器にしてみた。 vLLMをカスタムして、文章生成なしで候補用のlogitを直接計算。DGX Spark上で、複数項目の分類結果を確率付きで返している。 https://t.co/rhK5ygAuMK

never forget your favorite cup of coffee super proud of my submission for the all gas hackathon we've got: - @convex components galore (static hosting, auth v2, agent, aggregate, workpool, rate-limiter) - @firecrawl to scrape coffee roaster product pages, paired with a little jev action to classify the data - @openai's 5.6-luna in a convex agent to recommend your next bag of coffee by comparing yo

我给 tibo 做了个插件，判断 tibo 这条帖子有多大的几率会重置 codex 额度，使用 jev 模型来判断 https://t.co/yoJBCb10MC

fun weekend project: AutoJev. i was curious to see if i could train a competitive Jev-like model completely autonomously with a swarm of agents using our internal system. turns out you can get quite far! some details: - gave the swarm a devbox with an h200 gpu - the swarm is a mix of astra and fable; used sol and luna for synthetic data and filtering; ran for 20 hours - spent $3.1k in total ($1.9k

我重新优化了项目，目前已经速度已经达到了56秒。JEV真的太炸裂了。 https://t.co/XPMyp1EbXj

got tired of reply guys.. so i built a chrome extension that stamps them jev by @typesafeai reads every reply and hides the low effort ones want it? thinking of open sourcing it https://t.co/bwc9wU167S

如果你也是这两天被 X 时间线上的 Jev 模型刷屏、正跃跃欲试的 AI 实践者：我刚替你踩了一个价值 33% 周额度的大坑😭 先说我的结论：盲目把jev模型塞进 Agent 流程，很可能会换来翻倍的账单和更拖沓的执行。 为了验证 Jev 模型的实战表现，我把它接入 Codex，希望配合 Astra 做 Computer Use 下国际象棋 ♟️。我让 Astra 独立执行作为 A/B 对照组，跑完这轮实测，我苦笑不得：额度消耗多出2倍，用时反而更慢。 再看许多Jev的成功用例，我才发现，Jev 真正的舒适区，不在于像国际象棋这种，需要深度逻辑的推理环节，而在于快速思考的执行层。

I turned Jev into a directory finder for your saas. A founders DR jumped to 27. Mine 20. Free, no signups now. let's see your DR ↓ https://t.co/cA9bTWgBPo

1/ Jev has a chess rating now. 1023 Elo. Jev is TypeSafe's "System One" decision model. No search, no planning. You hand it a situation and options, it picks one with a confidence score. Built for routing and classification. Not games. So I built it a chess harness 🧵 https://t.co/0MN8P7druY

A model that can't write is routing my entire stack. Jev returns judgments, not prose. Task. Difficulty. Private. Needs web. Four probabilities, one tiny call. Then it picks the right model from a fleet. The demo's in the video. Judge. Decide. Route. https://t.co/Mhtz8MBjFQ

I plugged Jev into an AI marketing team. It read 700 competitor ads for 2.7 cents in 41 seconds, then gated every post before it went out, and sorted 800 DMs before I opened the app. Claude writes. Jev decides. The tools do the work. https://t.co/m9fPMqQpVp

That old M1 Mac mini has a job now. Running on Omarchy, I worked with Codex to build Jev Atlas — a source-linked map of 171 Jev projects, with search and a live graph of the ecosystem. Codex built it, tested it, deployed it through Hostinger MCP, and verified the public site — all orchestrated from this M1. This is starting to feel like what an AI-native computer should be. @maralcbr @CompleteSkep

Built a small experiment to understand my chess beyond “blunders” and “accuracy.” https://t.co/7bAuPHAGSH games → Stockfish for objective move analysis → Jev for recurring semantic patterns → code for trends and loss/win comparisons. Now the dashboard can show what keeps going wrong across games, what actually correlates with losses, and where I should focus next. Small beginning.

How fast is Jev? Can it keep up with Base's 200ms? 👀 I built a demo where you bet against Jev to predict Bitcoin's price the next second No sign up, no need to fund your account. It's all setup for you using account abstraction on Vibenet Try it out: https://t.co/v3bvCmGswu https://t.co/IaxHUzJWMB

Jevで50vs50のRTSゲームを作ってみました！ ・Jevが軍の全体命令＋個別ユニット命令を行います。 ・ユニット一人が目視できる敵の情報を定期的に情報収集→戦況を基に再度全体命令＋個別命令を行います。 瞬間的に１００人分の大量の情報収集＋１００人分への命令実行を行うという形です！ 結構見てて面白い✨

看完@NFT_Chen Laya × Jev 的贪吃蛇对比，我按照他给的源码库照着这个思路做了个「陨石穿梭」🚀 同一套陨石、同一套规则，双屏自由跑 30 秒：左边飞速穿梭，右边还在等下一步。 把决策延迟变成游戏画面，差距一下就直观了。 https://t.co/xQscBdtAbX

做了个工具，用 Jev测试我的 X 号能不能 90 天开通创作者收益。 Jev 的判断： · 换打法再试：64% · 投入产出：明显在亏 · 主要问题：投入太多、效率低（76%） · 离 X 收益门槛：还差 25 倍 · 再这样做三个月会撑不住：46% 难搞哦，链接在评论区🤡🤡🤡 https://t.co/Y57cXC3wg4

UE5で騎馬戦バトルみたいなのをJEV制御でやってみた。メタ視点で指揮官的な立場で、判断して指示を出している。今のところAI対AI。毎回戦闘結果が変わるので、眺めているとおもしろい。ただ、UE5のAI機能自体でやるよりも性能がいいかどうかは、まだ判断つかず。 https://t.co/2xClcLk1uF

https://t.co/uJzeLAzHtv 基于Jev制作的乐队指挥家小游戏 https://t.co/Qbb1uX8PY9

話題のJevを使って個人情報を一気に入力する機能を作ってみた。もう個人情報をちまちま入れることから解放された。ガチでAIの進化が凄すぎる。 遊んでる暇はない、AIに適応した人としていない人の差が、必ず顕著に現れる日がくる。世界が変わる。しかし、AIのしすぎで肩と腕がいたい。時間が溶ける。 https://t.co/oU28nsk8Cv

受启发也搞了一个，用 Jev 识别剪帖板信息实现表单自动填充 https://t.co/pvYnvmHBGA

基于 jev 做了一个 AI 帮你选的网站 https://t.co/S1PiRIkQyk

全網都在討論的新模型 Jev，是 TypeSafe AI 做的判斷模型：不生成內容，只從事先列好的答案裡挑一個，並附上機率。官方說它比常見大型模型最高快 193.6 倍，成本不到四百分之一。 我把它接進自己兩個真實工作做了 AB 實測。影片裡有它跟 ChatGPT、Claude 這類大型模型怎麼分工、兩組實測的結果，以及用之前先問的四個問題。

I know I know im late...but here is my JEV test. Classifying posts about Madrid fashion week. https://t.co/qjp2JlgldH

Continuing my JEV experiments: shopping by occasion. “Something tailored for work that I could wear to dinner afterwards.” The collection updates as you type, powered by live JEV calls. Credit to TSR Paris for the amazing collection: https://t.co/PBQMTisp0m Demo 👇 https://t.co/VVc4JUcCz7