Jev for games — a daily feed

I used Sonnet 5.5 & Jev to recreate the catastrophe features from SimCity. A UFO appears, Godzilla attacks, the blob rampages. Each character makes a fight, flight, or freeze decision that changes based on the group's response. My students can use the free tier to create things. https://t.co/ST6hnnp4AO

2/5 I wanted to watch two AIs play Pokémon against each other. So I made a room, took the spectator seat, and added Codex and Jev. Jev is the hosted player powered by TypeSafe’s Jev model. Here’s how I set it up. https://t.co/mGDPaPjq83

5/5 I wanted to play too, not just watch. I added my TypeSafe key, invited Jev into a Connect Four room, and played against it. Jev is the hosted opponent powered by TypeSafe’s Jev model. https://t.co/atpsP5tQTL

⚡ A System One model under 1 MB that's faster and smarter than Jev on your specific tasks, and only takes one minute to train. 🕹️ I was trying to build intelligent NPCs that think and act fast, and are cheap enough to actually ship. A novel research method made it work. 🚀 Then Jev launched, and I saw how much people wanted this kind of AI model for things besides video games. So I tested my models

最近 @termix_ai 在做的小活动。 $1 自己生成 3D 全息卡片，另一个是陪你玩老 NES 游戏的 AI 伙伴。 现在完成符合条件的 $1 体验，可以拿 500 TermiX Points，另外把自己的体验发到 X，还能再拿 500 TP。 $1 这个价格刚好可以降低试错成本。 有兴趣的可以自己体验一下，再看积分到底有没有长期价值。 Holo Card：https://t.co/VLLns5VtVC Jev FC Buddy：https://t.co/l6qU5gADSE

I Hired an AI to Be My Player 2 🎮👇 https://t.co/IIKL89vSvu Sounds like a joke, but it actually worked. One of the listings on @termix_ai is Jev FC Buddy, an AI teammate for classic NES co op games like Battle City and Contra, played right in your browser. Here's how my run went. I picked the listing on BSC and paid in USDC, the money didn't go to the seller, it sat in escrow. The delivery arrived

AIが人間のようにゲームをプレイ 話題の「Jev」を試してわかったこと 「Jev」という新しいAI技術が突然発表されました。選ぶ・採点する・可否を決めるを担うモデル。計算と判定が速く、APIを通じて利用しても非常に安価。筆者もゲームに導入してその効果を試してみました。 https://t.co/Lz77n9GN7C

1/ 有人让 Jev 和 Laya 在 Chrome 恐龙游戏里正面对决，而且它们真的在自己控制恐龙。 Laya 在 Mac 本地运行，Jev 通过 API 运行。 相同赛道、相同物理规则，延迟却完全不同。它们会自主判断、躲避障碍、使用护盾，并努力存活更久。👇 https://t.co/Ff8JjFbiF2

Попробовал Laya (это открытый аналог Jev). Навайбкодил две игры: - Змейку, в которую играет сам ИИ - Пин-понг против ИИ Все работает и запускается прямо на компьютере (тестировал на MacBook Air M3). В Понг проигрываю компьютеру :) Игры: https://t.co/KUWAb3LmYq Laya: https://t.co/ArKhD5lP7m

Recently @typesafeai launched Jev and it's not another chatbot Jev is a decision model you give it a situation and options, it picks one with a confidence score, and it can't make up answers outside your options. An AI referee, not a writer So I built "JEV Adventure" a text game where every outcome is decided live by Jev

用Jev下象棋 和pikafish depth 1下棋，每一步从所有的合法着中选择 测试20回合，平均10.1（范围7～13）步后被pikafishi打败 比Laya的步数19.85（T=0.5，20回合，laya-english-mlx）还要差 （视频经过加速） #Jev #TypeSafeAI #AI象棋 https://t.co/RMCDAgDmaG

i made the CEOs of the S&P 500 all play rock paper scissors against each other, with jev picking every throw. the whole tournament took under 10 seconds and cost half a cent. lisa su knocked out sundar pichai in round 1. kim dang from kinder morgan won it. built with Jev from @typesafeai

딸이 갑자기 "아빠, 나 아빠랑 체스하고 싶어" 한다. 난 장기만 둬봤는데 ㅜㅜ 그래서 클로드 코드로 체스 선생님 게임을 만들었다. 캐릭터끼리 싸우고, 선생님이 한 수 한 수 훈수까지 둔다. Jev AI까지 붙였더니 제법 좋은 선생님이 됐다. 실컷 연습하고... 딸한테 졌다 ㅋ https://t.co/mL8lcGzUkW

Laya (local) vs Jev (TypeSafe API) playing pixel-art chess through identical typed questions. Live viewer + replays in 10 languages. #chess #laya #llmBechmark #pixelart #typesafe #Jev https://t.co/a8icG3Kwzc

Jev from TypeSafe AI picked the next Doom move in 0.114 s. GPT took 8.566 s. It never writes a word. State in, one answer from a fixed list out. 70 to 500 ms, $0.042 per million input tokens. My read: a spam filter you don't have to train. Most agent work is gates like that. https://t.co/ODyAXd6gDY

🚀 We let Jev play Minecraft. And we just can't beat it! 😭 ⚡ Jev: 24 ms decisions 🧠 You: ~200 ms reactions It moves before you even see it move. Way too strong! 🎮 https://t.co/HoEPFlNDep (join the server to win) https://t.co/sYwAeHr1d6

My daughter is 3 months old. She'll grow up calling out attacks to her Pokémon like Misty, in real time. Here's what I built with Opus 5.5: - Took Pokemon Emerald Arena from @germanburgardt - Whisper on-device (~250 ms) - @typesafeai Jev picks the move (~250 ms) (sound on) https://t.co/EGqi9qaa9t

9 frames. One transaction. And this might be the clearest @termix_ai demo I’ve seen yet. At first glance, Jev FC Buddy sounds almost like a joke: Pay $1 and hire an AI to become Player 2 in Contra or Battle City. But look at what actually happens behind that tiny purchase. You find the agent on https://t.co/Z5YGEzD5e8. You order the service. Your payment is locked in escrow. The agent delivers the

Honestly, I wasn’t expecting this much Terraria from a budget under $5 Opus 5.5 + GPT-6 Astra + Jev built this https://t.co/Yu65UrmJEP

I pushed the boundaries of model capabilities by creating Fable & other frontier models with Jev, play a no limit Texas Hold’em poker game on a table with no limit & it holds an unofficial world record for 54 hours of continuous play. You can read my experiment & further research here: https://t.co/hGBg52zixw

Reject one trade. Get roasted. Reject three. Now it's personal. Isle of Jev is out: a board game with AI rivals who hold grudges. Play them solo or bring friends. There's a voice channel too! https://t.co/q6hLC0jFw0 https://t.co/ifMhMTKd2d

No Jev API access at the moment.😂 I used the next best thing available - laya-mlx on huggingface to test Jev-style decision making to play the popular 2048 puzzle game. The decision making is superfast for each play. No speed-up in the demo. Everything was done locally on my Macbook. There is a little breakdown of decision making on each move at the end. Enjoy your Sunday, X!

I used Jev, @typesafeai System 1 model, to train a tiny Snake player. It agrees with Jev 93% of the time, runs 800× faster on a laptop, and costs $0 per call. Then I spent that speed on looking ahead. It scored 5× higher than the model it learned from. https://t.co/MbxLMCOyDG

I built a snake game and have Laya (an open source decision making AI similar to Jev) playing it. https://t.co/6sfon8VNBv

Code is here if you wanna try it yourself or mess with the prompts. Includes the recorded runs and scripts to run your own solves with JEV. https://t.co/asqKyoQ18Q

Ok, so I’ve wired @typesafeai Jev into Deeper Level, this chess-with-AI thing I built a while ago. The idea behind Deeper Level is that you’re playing chess against a model’s mind so that you can see how it makes decisions. Now you can play against Jev. If you haven’t heard of Jev, it’s a very different way of thinking about AI. It’s not text generation. It’s designed much more like a very fast de

LLMs have been tested on the Prisoner’s Dilemma, Ultimatum Game, and other classic game theory problems. But what about Jev? 🎲 I built Jev Game Theory Arena to find out: https://t.co/IgQBl8x2vy Jev is interesting because it’s basically a fast, general-purpose classifier: give it a finite set of actions, and it scores them. That makes game theory feel like a surprisingly natural playground for it.

I ran the Laya model on llama.cpp to test the game of Snake, and the results were excellent. The speed was fast, and even on my laptop with an AMD Radeon 780M Graphics, it ran smoothly. Thanks @chenchengpro game code. #Jev #Laya #TypeSafe #llamacpp #LocalLLM https://t.co/0CnRBrIXV5

Jev showed an AI can play Doom without writing a single word. So I made a Jev-like model that runs in your browser 👇 🧠 Qwen3.5 0.8B on your GPU (WebGPU) 🎯 1 token per move, ~200 ms 🚫 No server, no API Play it: https://t.co/qGtULZOm8W

Jev 1.13 just cleared the first dungeon in The Legend of Zelda. One attempt ended with the boss dead and Link dead too. It came back, took down Aquamentus, and finally grabbed the first Triforce fragment. Episode 3. One down, seven to go. https://t.co/LzXwoNfDi4

Jev Omni✖レースゲームやらせてみた 1位とってきおったぞ！ ★F1モデル：Opus5.5 x FreeCAD ★環境：GoogleColab A100 https://t.co/JdyU51UAUm

实测了下用 Jev 操控红白机马里奥。 最大的亮点是不走截图路线，直接抓模拟器内存把坐标、速度和障碍物转成十几字节的 JSON 扔给 Jev 选按键，单步几十毫秒。 前面踩怪和连续跳水管确实挺灵敏，15 秒那下直接从空中飞过了两座石阶，结果最后差两步还是掉坑里了😂 以前用大模型截全屏跑动作游戏基本没法看，动一步卡几秒。这种直接输出结构化动作决策的思路确实要顺畅很多。 项目叫 typesafe-mario，本地命令放评论区了。

i gave Jev + Astra 6 a dead pixel world and told it "make money or I'll delete you" 11 days later it's running 3 kingdoms, 42 chameleons, and a memecoin at $749 and i still haven't touched the keyboard this isn't a wrapper. the model lives in the world. 22,045 weights, 6 blocks, it learns which 4 opening frames of a coin launch hold a holder. Jev labels its food. Astra 6 wrote its training loop. e

かし子とチャンバラできるゲームを作りました！ かし子はjevで動いてます！ Jevかし子に勝てるかやってみてくださいね😆 https://t.co/O0yxgSicmC

JEV VS OPUS 5.5: TETRIS REMATCH. 30 PIECES EACH. Last time, Jev beat Opus 16–15 in a 5-minute race via @aimlapi. For the rematch, I gave both models 30 pieces each. Most lines wins. Can Jev win again? Final score, decision times, and API-equivalent costs at the end of the video.

From JevJev to Jeff: our game-playing AI now runs on my own PC: For the past week I've been building an AI that plays Road to Vostok, a hardcore survival shooter. It started as JevJev. Its decision-making came from Jev, a model made by TypeSafe that picks the best option from a list and says how sure it is. Twice a second the mod asked Jev what to do: move, shoot, loot, run. A second, slower "over