All builds

Jev Plays Pokemon update. $1.21 spent. 8,000+ decisions. Jev has beaten Brock and gotten the first gym badge. Now leveling up inside Mt. Moon. Follow along; https://t.co/4Gj24SpIfR https://t.co/Csr7w8o1lF

I recently got access to jev from and built this, because I hated the hacker news front page. everyone gets the same stories in the same order. someone else already decided what matters, and it is never quite what matters to me so: six sliders. technical depth, drama, practical utility, ai slop, novelty, career relevance pull ai slop down, push novelty up, and the list rearranges itself. max out d

JevでNext Edit Suggestionできた!!!!!!! もうみんな多分存在を忘れてると思うけど、若干面白いかもしれんwww https://t.co/YQutrEhlqT

Jev can play and beat the hardest gamemode in BTD6 (CHIMPS) with no outside help. - realtime, fast forwarded - rounds automatically start with no pause between rounds - self learning by looping game attempts and feeding the data back in - this is the recording of attempt 38 Astra/Fable cannot do this.

Making a smart color picker with jev. type in anything and get a color https://t.co/y1sf0ZcZhr

Here is Jev from @typesafeai classifying 300 full chat sessions from our custom @vercel eve agent by failure reason. 323 sessions, 2M tokens done in 14s and costing $0.08 https://t.co/VRQKvuxvJV https://t.co/OHmFW4XTtc

I gave Jev a robot arm. Goal: red cube in the green box. Catch: the cube was out of reach. It grabbed a hook, dragged the cube closer, put the hook down, and finished the job. Tool use was never programmed, the model just decided. Fast enough to run a robot 👀 @CompleteSkeptic @typesafeai @dotpem

i have created moderation guard based on typesafe's jev its fast and accurate live demo you can: https://t.co/FquTJtWCzq https://t.co/PyNNZGu0iW

20 comments. 1.48s of API time. $0.000441. I wanted to try Jev from TypeSafe AI, so I built a moderation demo. First idea that came to mind. Next: testing AML alert triage and data labeling. If the quality holds at this speed and cost, there's serious potential. https://t.co/grCKDMmQNT

Spent an hour on YT livestream today with the Jev model via the typesafe API. Thanks to typesafe team for helping out with credits. The main use-case for this LM is to predict context aware decisions when the decision space of the problem is known apriori. The biggest downside is that you CANNOT generate new tokens - it only works as a classifier + confidence-predictor. You cannot use it even for

Deux choses absolumment incroyables s'exécutent sur mon laptop aujourd'hui. Le cerveau d'une mouche... et Jev, un nouveau modèle de décisions calibrées (RLCD) conçu pour choisir une action parmi un ensemble fini. Application d'évaluation: Le Guet, un pilote autour de Genève. On lui donne un écran, des actions possibles et un objectif en langage naturel. En une requête, Jev décide quoi faire ensuit

I built a startup judge powered by Jev. You describe your idea. Jev makes a bunch of fast, typed decisions about the problem, demand, monetization, distribution, differentiation and more. Then Kill My Idea gives you one answer: KILL IT 🔴 FIX IT 🟡 SHIP IT 🟢 No startup therapy. Just probabilities.

We implemented the new @typesafeai Jev model to Luo and I tested it with a help center classification use case. User types in their question or issue and in realtime the system analyses what the message is about and which help center article fits best, if any of them. I compared speed and cost to using Haiku 4.5, Anthropic's cheapest and fastest model. Using Jev is 2-4x faster and ~25x cheaper.

We ran an experiment fixing bugs in a Snake game using Gemma-4B paired with Jev. The setup was built to isolate executive function from code generation: vanilla Gemma had to decide every action and write the patch, while Jev handled routing through a four-part evaluation checklist (next action, evidence threshold, loop detection, completion) and only invoked Gemma when an edit was required. Across

I've made a Chrome extension with Jev from @typesafeai that reads my feed and hides some posts according to some filters. Because the Jev is so cheap and fast, it costs almost nothing to run. Let me know if you need a Jev key to test it https://t.co/PKX7OxzVrb

Je suis HS dans mon lit 🤧 mais j'ai demandé à mon agent de faire un poc rapide pour https://t.co/BAt2V3Hb65 Les applications sont infinies ça ouvre un nouveau monde des possibles. Par exemple ici on a créé un système qui permet à l’avatar de prononcer en direct un stream en y ajoutant des expressions. L'app découpe automatiquement en petits passages cohérents. Dès qu’un passage est complet, JEV an

Book an Uber in 20 seconds⚡️ I finally got access to Jev, and the first thing I did was give it a phone. It’s blazingly fast at executing actions. At this point, the only bottleneck seems to be the UI animations themselves. https://t.co/hNrJK60eVW

@typesafeai のJevにテトリスをさせてみました。API返答は東京で250msぐらい、ほぼリアルタイム。 渡した事実を読み違えないのと、confidence がちゃんと機能するのが印象的でした。柔軟で汎用的な分類器としてかなり実用的です。 マルチモーダルなど発展の余地が大きそうです。楽しみ！ https://t.co/3tZC6PuS91

This is cool. I used Jev to improve my custom memory system. Results: → 94% fewer tokens used → 2–3× faster memory retrieval This was just a quick test, but I believe it opens up so many new opportunities. https://t.co/Xppb20drpY

昨天出的 https://t.co/6eiMFaJTrj Jev 模型 适合高频决策，那岂不又是适合打游戏？ 之前用 GPT-6 Astra 代打杀戮尖塔2 能力强但速度慢。 我刚实测用 Jev 打，行动思考只需 0.7秒， 画面我都没看清它就操作完了。 这超人类游戏速度， 着实又让我震惊瘫坐了😅 语言很难描述这感觉，看视频⬇️ https://t.co/y9Nu78Hnv4

10分で作ったやつ、Opusと5時間いじってこうなりました。 trueX（トゥルー・エックス）は、X（Twitter）タイムライン上の投稿をリアルタイムで解析し、「公式発表」「未確認の噂」「個人の感想」「ネタ・デマ注意」をスマートに色分け表示するChrome拡張機能です。 文章を生成しない次世代AI「TypeSafe AI (Jev)」の超高速判定（200ms台）により、ブラウジングを妨げることなく、タイムラインの情報信頼度を一目で把握できます。 【主な機能】 ⚡ 超高速リアルタイム判定 タイムラインを流し見しながら、各投稿の事実裏付け確率や確信度スコアを200ms台で瞬時に判定・表示します。 🎯 スマートな4層コンテキスト分類 ・📢 公式発表・事実（青）：確定した一次情報や公式ニュースをハイライト ・🟡 未確認の噂・リーク（黄）：未確認情報を「デマ」と決めつけず、注意喚起で表示 ・💭 個

Breaking: Jev can detect AI misalignment I had it scan through real ChatGPT conversations to identify cases of misalignment, and it's blazing fast. It flagged about 6% of chat logs from a small sample. I believe this will be a great way for unbiased 3rd parties to conduct safety research on frontier LLMs.

been experimenting with this: what if the SQL agent only gets the schema it needs? added @typesafeai’s Jev to select relevant tables + columns before Astra writes SQL. @hackgoofer you asked for production workflows... here’s a prototype tackling schema selection in text-to-SQL 😆

I built a New York JFK airport manager simulation with Jev in less than an hour inside Agent Zero. This model is insane. https://t.co/GLjQdWz1KF

Jevで次の文章予測判断させるロジック作って、 判断特化AIのJevで、AIチャットできるアプリ作ったんだけど、まず爆速だし、入力は$0.042 / 1Mトークンだけど、出力が無制限で無料でした。 出力無料のAIチャットってなんだこれw 「え？これ作れちゃっていいん？」ってなってる。。。 #Jev https://t.co/h5LDG2tPpN

flysim is a foraging world where a fruit-fly has to find sugar and dodge a predator, and you swap out who makes the decisions. I put two decision layers in it and ran them at the same time. Same world, same body, same predator — concurrent, not lockstep. Jev by @typesafeai lived 30 ticks in 13.2 seconds for $0.000834. gpt-6-astra took 101.9 seconds and $0.121280. 145x the cost. 7.7x per decision.

J'ai testé Jev (@typesafeai) sur plusieurs trucs que j'ai au quotidien : - soumissions de formulaires de contact : sur mon site https://t.co/qqMSeLGAsx j'ai un formulaire sans champs "sujet" pour des questions d'UX (la friction tout ça tout ça). Là, typesafe peut les catégoriser pour moi en fonction des messages reçus - pareil pour assigner des tickets Jira : j'ai testé typesafe sur une catégorisa

I built near-real-time A/B testing of TikTok/Instagram hooks against a 100 rich personas using Jev from @typesafeai! Stuff like this was super ROI negative in the past. A panel like this meant $1-$5 of cost per run at this scale, plus very slow. Now it costs just a few cents & is real time! This is a brand new world! This is a new ChatGPT moment. Great job @CompleteSkeptic and team! You guys cooke

I got access to Jev by @typesafeai today morning and I built a cool use case for it Introducing DiffJury - simply paste any public PR link and Jev tells you immediately if it's safe to merge or does it require review ✅ 🔗 Feel free to try it out here - https://t.co/XxcZxIZJDK Imagine Jev being able to tell you if you should merge a PR with grounded context of your codebase. that's what we're buildi

よっしゃ〜！！！無事セミナー実演で、DevinでJev使って、アプリにスピーディな意味検索機能を作れた〜すごいなDevinもJevも🔥☺️ みなさんご参加ありがとうございましたー！！！🙌 Jevの使い道すんごいあるなこりゃ！✨ #Jev #Devino DevinでアプリにJav入れた実演解説のアーカイブを1000円で買えるのは明日の朝8時まで、ぜひこちらからどうぞ👉 https://t.co/xwK71r2Ou3

I let Jev from TypeSafe draw a picture pixel by pixel and compare it to Claude Haiku. Just a tiny bit faster. https://t.co/6WwftF0U7Q

Built a tiny island demo with @typesafeai 's Jev. Six objects, three robots, one ferry with a 7-unit weight limit. You type what matters. Jev chooses the cargo, assigns the crew, and decides when to sail. The pirate hat was my decision XD https://t.co/nthKgVQht0

Jev by @typesafeai in action! 20.7k YouTube comments classified in 2m 27s for just $0.20 - p50 319ms, p95 556ms per comment - 140 comments/sec - sentiment + emotion + intent + spam/toxic, each with confidence mostly Apple WWDC videos and results are kinda funny: 43% negative #1 intent is criticism (5.1k) my classification rules are probably not ideal... but you can tweak them and rerun the whole b

got Jev to review my PRs. ~200x cheaper than Claude and it answers in half a second 6 real PRs in the video. $0.00007 each. 1,000 PRs = 7 cents vs ~$14.50 on Opus 5 paste a diff → ONE call to @typesafeai → 14 typed checks come back as probabilities: hardcoded secret, sql injection, touches auth, deletes tests, breaks api, migration, debug leftovers, does the description actually match the diff, bl

i made jev to play a trading quiz @typesafeai it gets given btc chart data from a random window between 2017 and today, then it has to predict the direction after n candles from the last candle it was shown. available actions are: buy, sell, skip (off for now) you can watch the activity here: https://t.co/AdxKUnvu2j

Jev from @typesafeai is really cool. I made a battlefield demo and each character makes their own decisions with Jev. Watch the video or watch it on: https://t.co/RQEqkoLVss https://t.co/p17KLZWtmO

@typesafeai 's JEV is actually pretty neat. I compared it against claude-haiku-4.5 on the same payment transactions, with 8 concurrent threads evaluating each stream. JEV was noticeably faster and feels well-suited for real-time classification. Accuracy is always debatable for any model, but infra can be built around confidence scores. AI slop -> https://t.co/E8EbJqRdXD

Built the ultimate syntax highlighter using @typesafeai Jev (i mean why not). It can highlight any language, even one that I made up.

@typesafeai 's Jev is awesome. I've used it to create a custom/dynamic MCP server framework that can turn any site (including ones requiring auth) into an extremely fast MCP. * mcps isolated in docker instances * Jev drives headless chrome * agent skill - just instruct as to which site you want to mcp-ify and it'll do the rest (maybe) * inspired by @gregpr07 's jev-ultrafast https://t.co/Qts00CR1Y

I gave a vampire, a golden retriever, and 10 other unqualified characters a vote on your bad ideas. Powered by Jev by @typesafeai: 36 typed judgments + probabilities. One call. ~400ms in early tests. No generated text. https://t.co/0DkzIRxIHR