X posts

完全睡不着了....这个视频让你完全看懂JEV的恐怖能力 我实际演示了一下，用 JEV 和 DeepSeek 做同一个任务：从当天的海量实时新闻流里，为 15 个品牌快速匹配有没有适合借势的公关热点，并打上结构化意图分类标签 结果是：28 秒内，Jev 狂刷完了整整 428 条数据，并完成了全部的目标分解与分类；而同一时间跑在同样任务上的 DeepSeek V4.1-flash，才刚刚完成 6/428 条 速度被拉开了几十倍，答案甚至在你读完第一行字之前就已经返回了.... 为什么会产生这么恐怖的差距？底层逻辑是： • DeepSeek 这种通用模型本质上是在“逐字写文章”： 哪怕你只让它做个最简单的“是/否”判断，它在后台也必须一个词一个词往外推，硬走一遍漫长的生成流程，延迟按秒起步 • Jev 官方定位是“System 1（快决策）”模型，从根上就根本不会写字： 它彻底抛弃了逐字吐词，

Can Jev escape a maze? Turns out it depends how you write the map. When given as a 2D grid, “wall above?” is worse than chance. “Wall to the right?” is 99.9%. Rotate the page and every side becomes 1.00. It isn’t seeing the grid. It’s reading the next character. So I gave @typesafeai’s Jev the maze as lines of sight and named places, not as a map. Notes below.

love jev, but upset it - isn't open source? - it lack vision capability? We fixed all of that, introducing https://t.co/DlNeKo2k5e A fully open source library which takes any HF model and Jev-ify it, with an API endpoint Now on github, and live in production at @FeatherlessAI https://t.co/dFKBbR411y

ローカルJevできた。確信度ないけどｗ 40msで返事返ってくると不安になるけど一応GPUはピクっと動いているのでLLMの出力にはなっている https://t.co/8eEtZ0CpSU

the obligatory AI plays Mario, Jev version Jev is TypeSafe's System One model. u send it state and a typed question, it sends back a choice with probabilities. so every move here is one question: run, hop, jump, wait or go back it reads the game from NES memory, no pixels. 193 calls to clear 1-1, abt 300ms each right side is the real request and answer for every decision. the game is paused while

Jev vs GLM 5.3 at chess! Results: ◾ GLM 5.3 won by checkmate in 29 moves ◾ Jev: ~0.3s and <$0.0001 per move ◾ GLM 5.3: ~5.8s and ~$0.008 per move ◾ The whole game cost 24 cents My main takeaway is that it's often useful to use each one to their strengths: ◾ Fast, well-defined classification → specialized models like Jev ◾ Classifications that need reasoning or lookahead → LLMs like GLM 5.3 ◾ Real

Tried Jev for this use case : On one side: a JD → extract the most important requirements. On the other: upload multiple resumes → see which candidate fits the JD best. Jev was surprisingly fast at this. I will test this model on a few other use cases I have in mind. https://t.co/b0IkQ3zDjM

I got tired of choosing between X's firehose and someone else's idea of relevance, so I built Your Signal. It uses @typesafeai's Jev to score posts already on screen, then applies my rules locally. Reversible, BYOK, no telemetry, open source. https://t.co/1dNRESvT5J https://t.co/sYOVB3eIoy

**jev scroll break** took a stab at making a design skill inspired by @anshuc dream loop skill, and made taste loop! tested it on my portfolio cuz it needed an update and holy it looks much better now https://t.co/6z0lsmyHt0 https://t.co/lgspHpuKrA

DiffusionGemma-as-Jev (aka djev) running near-real-time vision detection from a mobile phone using its native vision tower. Please don't fall down the stairs! https://t.co/C9kBCD6irx

I built a game were you play against a sentinel in two modes: 1. Default: Sentinel runs a fixed chain of if/else rules. It follows every one of them and still wanders the map looking for me. 2. JEV Live: hands the Sentinel's tactical call to JEV. -> It went and covered the exit. 💀 Same game, same available moves. Only the decision logic changed.

@typesafeai @CompleteSkeptic lol Jev is fun got it to solve a bunch of games at scale, Sudoku took a bit of time though https://t.co/hLrF09riQ1

Jev x GPT-6 Astra x H3 Max I created a complex decision making game with Astra. It is about keeping the city on a whale alive with decisions. - GPT 6 Astra designed the world - Jev chose the actions - H3 Max Turbo on fal turned one decision from each round into video All of the Jev decision making and video generations took only 5 minutes for 264 clips

Jevmaxxing on Notte browser sessions ⚡️ > https://t.co/ZokbHG5WxE Type a task, hit start, and watch Jev @typesafeai pick every actions super fast from the page's action space. > new action space every step > Notte cloud browser over CDP > step-by-step replay of every decisions Try it yourself! ▸ https://t.co/ZokbHG5WxE ▸ https://t.co/uGO7cLo2st

tested Jev as a real-time robotics policy in MuJoCo it struggled at first, so i split each update into two calls: decide what to do next, then decide how to move the arm and gripper Jev doesn’t accept images, it gets simplified geometry and contacts as text here https://t.co/iKK7jcrrpO

Also been playing with @typesafeai Jev. Insane. A pile of apps suddenly become possible. What a time to be a builder. Sharing experiments here. First one: Keystroke oracle / predictive launcher A normal launcher ranks by aliases, fuzzy match, and habit. Jev reads intent. Type “the pdf I just downloaded” and the newest PDF is already the top hit. Full confidence on every keystroke. About 100 ms.

another demo for jev for ~instant variable generation https://t.co/ddjvvHXpai

Jev + Higgsfield = solved auto-routing for genAI models. In this demo, @typesafeai’s Jev evaluates prompt and picks the most fit models for video and image generations on Higgsfield API. https://t.co/MyTRdqlZ3o

Jev + OpenCode + Browser Harness + Sanbox = 🔥 Watch how OpenCode + browser harness visit a tetris game website, read the control instructions, use gpt-6 to build a game controller for Jev model, and then hand off to play the game. This is all a single shot prompt with no special skills or tools. Jev model chooses from predefined outputs. In this setup keyboard based gameplay actions were not inclu

I used Jev to build the fastest SEO/AEO audit tool it took it 16 seconds to audit my full website It goes through every page on your website, analyse it checks it's html code, content and gives a full prompt to fix them too. It's free and available on Bottally now. https://t.co/oYTFCWYjUY

Jev added more precise camera controls and ~230x faster planning to an editing tool I built on top of @mintdotgg! Jev converts natural language to precise camera movements (ie spelling ASTRA) by answering a few calibrated multiple-choice questions (move kind, plus direction and magnitude band per camera axis) By creating keyframes it emits a structured trajectory (time/azimuth/elevation/distance)

yes / no jev demo https://t.co/cINZEnqWFm

Applying @typesafeai 's Jev to robotics & swarms. JevDuck (https://t.co/suJgqvf7pv) connects Jev's typed decisions to the official Microduck simulator. MuJoCo WebAssembly computes the robot dynamics in the browser and the original ONNX policies control the articulated joints. Try it here https://t.co/r8DToTfmcg

Made Jev fight OpenJev in a first person shooter. https://t.co/wSB3psnvwB https://t.co/vwewLu682T

I used Jev to give a tiny civilization six minds. Then I played GOD. Kindness earned trust, Tyranny sparked rebellion and an apology didn’t convince everyone. Jev decides how each faction reacts and the simulation brings it to life. Try GODMODE in replies ⬇️ @typesafeai #jev https://t.co/qRiNOIylwa

Got @typesafeai jev-latest access today So I made jev play the dino game against a gpt-5 model and a real human and the results are shocking!! https://t.co/YGnubYe7Ps

made some time to test jev from @typesafeai used theMealDB and sent each recipe to Jev to score on five dimensions: quick, cheap, healthy, impressive, and beginner-friendly, while filtering out desserts, sauces, bread, etc. The whole run took just over a minute and cost $0.0455. I ran it twice and got slightly different rankings. Close results swapped places over the runs. So you either need to us

Was curious how jev performs for e2e testing on a real android device. Both started with the same prompt and 15 test steps. Well, jev was ~14.8x faster than deepseek + vision navigating through wikipedia. https://t.co/xLW13tBISH

jev is bloody INSANE we gave it a library of over 11 million videos on t1ktok & 1nstagram and asked it to find viral "bloom nutrition" content. in 20 seconds, it gave us over 384 videos, breaking down the hook, format & angle that made each one go viral. (and it only costed $0.09 in token usage) jev finds, watches, analyzes and breaks down viral videos and creators, then writes briefs and scripts

Jev can control a swarm of 15 autonomous drones in real time! I built a simulation to test it navigating an asteroid field: • Decision latency: < 300ms • Survival rate: 100% (all 15 reached the destination) Built with @typesafeai by @CompleteSkeptic https://t.co/4xbduOkArc

Now it's my turn. Jev sorts my inbox very quickly and accurate and also let's me add custom filters? I was working on my own mail client for private use for mac and linux and today i wired it up to use jev to sort my mailbox as i need it. Also in there are AI replies (not jev, use any model you like gemini is good for this) and all works with an Openrouter API Key Who wants to try it?

I built a real-time filter for X with Jev as you scroll, it spots engagement bait, rage bait, and promo posts, then throws them into a trash bin. and it's open source so you could BYOK ⬇️ https://t.co/UyAjHiuwP1

JEV is INSANE 🔥 we gave it 1,873 backlink outreach emails to see which ones are more probably to get a reply (0-100) for each one it analyzed: → website and article relevance → how natural the proposed placement felt → contact message fit → personalization quality → friction of the request we then ranked the entire campaign and prioritized the emails with the highest probability of success this is

What does it cost to sort 1,700 emails with Jev? I gave JEV each full email object: - Subject - body - sender No special changes. Jev sent back 4 values for each email: - Category: work, shopping, finance, security - Priority: low to urgent - Spam score: a percentage - Reply score: how much the email needs a reply from me One user wrote that their account had a violation. JEV gave it a 90% reply s

The ideas have been flowing the last 48 hours. I built a way to load and classify YouTube comments with Jev to understand which ones need my immediate attention 🎉 https://t.co/cKOFKv1ZjX

@CompleteSkeptic Jev prompted me to explore this with open models. I built a small Python package for typed decisions using local Qwen, and tried it on Tetris alongside Jev. Code, traces and limitations are public. https://t.co/NeHEWQ8qBQ https://t.co/IZJYhsdmpM

I turned Jev into an image model…? Presenting Jevinci! https://t.co/8rO1mBA1D9 The paintings are pure JS, no AI image-gen. Jev predicts every pixel's color in parallel and we paint them. The more confident Jev is in a prediction, the wider the brush. (sped up 4x) https://t.co/jWkC0pB6uj

I built a bot that plays a puzzle game by asking an AI before every tap. Each tap you see is a live API call. This is Jev, @typesafeai model. No prompt, no generated text. My code lists the legal taps, Jev returns a probability for each, code taps the best one. 15 puzzles, 190 calls, ~240 ms per call, total cost $0.0298. (Most of that was the last hard puzzle) Why so fast and cheap? A normal LLM a

Weekend teaser: Jev vs Opus 5 The task? Win Minesweeper. ⚡ Jev: 24s | $0.003 🐢 Opus 5: 2:19 | $1.09 5.8× faster. 363× lower cost. Next week, we'll release a demo so you can compare different models and see how they handle your own board. https://t.co/quQOzli57l

i put Jev by @typesafeai inside a pigeon. it has one job: find jargon in your writing and take a dump on it. https://t.co/us8w5LNjdz