Jev for classification — a daily feed

Trained a Jev-style classifier on @huggingface Jobs for ~$1.50. It's a 194M GLiNER2 model that suggests task tags for any Hub dataset from its column names and first row, and returns a label with a probability. Zero-shot, GLiNER2's first suggestion matched an owner's tag 10% of the time. After 17 minutes of fine-tuning: 69%. The fine-tuned model runs on a free CPU in about a second. Owners' tags a

Jev all the things?? 🔥 built a CLI over the weekend that classifies an artist's entire discography by theme, mood, and lyrical complexity, using Jev (thanks @typesafeai) it then renders it as a terminal dashboard ran it on Nirvana. 52 songs, 1989–1993 pipeline is simple on purpose: → MusicBrainz resolves the artist + pulls the discography (free, no key) → https://t.co/w4qk26vniP fetches lyrics per

Dopamyn + JEV vs without JEV. Same crypto account tagging job. ~20x faster and cheaper with JEV. https://t.co/fhd0acgxbE

We tested Jev, the new AI decision model, on tagging notes. Its accuracy lagged, measured by how often it applied the same tags people had actually used across several open-source, publicly available note repositories. But its speed and cost were vastly superior to the traditional LLMs. More use cases to come…

BlockRun × Jev ⚡ Introducing SignalDesk. We gave Jev 100 real X posts. It made 300 intent + product-fit judgments in 3.9 seconds. Find people asking for alternatives, see if your product fits, and get a reply draft to review. Open source. Demo below. https://t.co/ekndVkcyDb

Jevで国会を仕分けてみた。 予算委員会の質問 2,424件を、分野×性質で自動分類。 1件2秒、生成なし、確率つき。 質問の文字数の 81% は野党 ・答弁の 25% は総理 ・内閣提出の法案は 64本中64本が成立、否決ゼロ。議員提出は衆法 14/38、参法 1/21 ・総理の答弁で「差し控え」が出るのは、野党の質問に 4.5%、与党の質問に 0% ・質問が触れた話題は、中東情勢 13%、原油・燃料 9%、消費税 4.5%（6月13%→7月28%）。地震 1.4%、水害 0.6% 出典は国立国会図書館の会議録API。

imagine all your tiny cars parked quietly in your pocket... so I built this for my hotwheels' collection with Jev by @typesafeai - Jev reads your query into intent + make + colour and scores every car 0-100% on how well it fits - Type the name or the serial off the back... Jev matches it against the whole collection, works out if it's mainline, silver or premium and says buy or skip against that c

to montando uma plataforma interna de satisfação pros produtos da DOTI e o primeiro dado que apareceu foi sobre a gente metade das mensagens dos grupos de feedback alpha era do próprio time. dev testando fluxo, gente respondendo dúvida, suporte interagindo... fiz a seguinte mecânica: as mensagens dos grupos no zap caem direto num banco e o jev classifica cada uma (bug, feature request, elogio, dúv

Qwen3.5-4BのJev風の判断器LoRaの学習が終わっていたので、同じ分類をやらせてみた。 データセットを追加した都合、確率の出方が凄く良くなった。個人的に使う分としては十分だろうか。 https://t.co/XhENcVEtKZ

今話題のTypeSafe AIのJevに映画のカテゴリーを分類させてみました。ソースデータは個人ブログに書き溜めた私の映画の感想です。感想にカテゴリーのキーワードをそのものズバリ書いている場合もあるのですが結構な精度で当ててきました。 https://t.co/q4KHjuFYjB #Qiita

Попробовал хайповый Jev для фильтрации данных. У меня есть задача - собирать посты о вечеринках из социальных сетей и выдавать релевантные редактору. Jev оказался для этого идеальным, быстрый, дешевый, доступно из России https://t.co/jLQW9HPOpJ

先週、自分がXで送った返信について、AIに書かせたものだと指摘を受けました。 理由は文体ではありませんでした。 構造です。 2文とも、相手の投稿の言い換えになっていた。 新しい情報がゼロだった。 先週出た、TypeSafe AIのJevで、自分の返信を採点してみました。 文章を書かせるモデルではありません。 判定だけをさせます。 文を渡すと0から1の数字が返る。 聞いたのは1つだけです。 「この返信に、自分が実際にやったことが書かれているか」 指摘された返信は、別の質問で0.85でした。 「元投稿の要約に留まっている」。 9本中で最高値です。 AIも同じところを見ていました。 過去の返信を8本、同じ質問で採点させたら、3つに分かれました。 経験型 0.72〜0.92 自分が実際にやったことを書いている。 意欲型 0.17〜0.18 これから試したいことを書いている。 共感型 0.06〜0.

The interesting question is no longer how smart the model is. It's which model it calls. Last week I asked Claude to sort the Hacker News front page against my interests. Claude didn't do the sorting. It wrote the code, then handed each headline to Jev, a small decision model from TypeSafe, with one narrow job: answer six typed questions about this headline. 100 headlines. 600 decisions. Three sec

OpenAI API 無料枠用につくった、右クリックとサイドパネルからAIに質問できるChrome拡張機能に、TypeSafe AI 「Jev」入れて判定させるテスト とりあえずなんか判定してくる https://t.co/YVQrsCCpKP

また、同じような処理ができるプロンプトを作成して、LLM Recipeとして実行をさせた所、53秒と17倍も高速に処理をしてくれることがわかりました。すげー!😱 #Jev https://t.co/s70K8F3uc4

Qwen3.5-0.8Bの方のJev風の判断器LoRaも再学習できたので試してみた。条件整えたら流石に爆速だな。（先頭900文字に限定したニュース記事の分類） https://t.co/i4scizQ4js

Brio mode in Colibrì Give it context + a closed list of answers. It returns the probability of each choice + entropy (how unsure it is). Zero generated tokens. Inspired by TypeSafe’s Jev — but on 9 open model families, running on your hardware. https://t.co/JUy6jAtaHH

Feed Lens 审核过了，终于能直接从 Chrome 商店安装了 🎉 给自己的信息流加点自己的判断。欢迎试试👇 https://t.co/tBmqky3iSN #FeedLens #Jev #TypeSafe #AI

Jev使えるようになったので、JDLの公開PDFをDLして分類するスクリプトこさえました。 https://t.co/J7TVKok3mn 全ラウンドの組み合わせとか、リザルトとか逐一手でするのちょっと面倒なのでこのスクリプト使えば一気にDLできまっせ。

Jev is hella nice for categorising my transactions in my personal budgeting software. Great for single/batch processing. Also shout out @EffectTS_ with their latest ai release with Jev. Made it easy to implement into workflows. https://t.co/R6ej9ivTW6

first thing I built: an LLM evaluator normally you'd prompt another LLM to "rate this 1-10" and pray it's consistent with Jev you just ask for a score and it returns 0.73. no parsing, no vibes https://t.co/ppbOd73PDR

We recently added @typesafeai's Jev model as a native capability into Luo workspaces. Whenever the workspace tasks or features need to classify information, it'll process it with Jev for faster and cheaper processing. https://t.co/TJm7SZQCKV

retired the good and old 2D view of my obsidian vault to something I actually enjoy looking at. Jev runs the classifications and the only thing I really need obsidian for now is the plugins.... kinda crazy ngl https://t.co/VPnM3k4tP5

@CtrlAltDwayne BTW, I have been working on model orchestration, I have done one plugin with MCP using Jev for difficulty classification of tasks, hooks to correctly handle all planification process when needed and track permissions of swap agents, and skills to control it. I am happy about it. https://t.co/4F5EhtYP9C

This might be the most useful thing I’ve built with Jev so far. A Chrome extension that quietly analyzes what I’m seeing and classify them as Opinion, Engagement bait, etc... So clean you may think it is a new feature on X :) https://t.co/POUtFhxGkD

I fed Jev our mobile session data. It flagged every frustrated session in seconds. Not gonna lie, I was pretty excited when I came across Jev. This is an attempt to use it to classify https://t.co/NcVmDAL3l5 sessions into healthy, degraded, or frustrated based on a set of weighted signals: - User-visible failure: failures the user actually sees - Failure in critical path: blocked on login, checkou

Given a use case with structured data, clear instructions and deterministic outputs Jev is over powered. I'm using this in production today using the @AskVenice API for my fantasy tennis platform @brckt_io. Come check it out for the next tournament which starts in a few days! https://t.co/kRHW3LJDLD

Jev とは何か — ブログ57記事のタグ付けを任せたら、正誤が付かなかった https://t.co/ZzdCRAsuOP #llm #evaluation #ai

Hooked jev to Claude via MCP to map an Angular base module (8 classes, 6.7k lines). Rather than guessing, Claude routed the decision via jev_route_task across 4 paths. Outcome: proceed_fast (0.34) edged out block (0.26). Pure visibility, zero guesswork. https://t.co/jyHeeMRN3r

Jev is now categorizing my LinkedIn feed. Live, while I scroll. https://t.co/ltMaPpz8J7

I had to do it. Had to ship Jev to prod. On a serious note, we were previously classifying restaurants with code and with LLM judgement. Naturally a perfect usecase for a classifier like Jev. Cut the "bad apples" before LLM saw them, and runs became much more "deterministic" https://t.co/NS9evvm2Wn

Jevで驚き屋判定できるようになったぞ！！ https://t.co/t7ayOwglEJ

Jevでノートの自動タグ付けシステムを作った🎉 Obsidian使っている人はぜひ👇 https://t.co/qAUSkuaVpD

He hecho una web para saber si tu próximo post es un banger o una mierda Lo escribes y Jev te pone nota Este post ha pasado por ahí, si flopea le echamos la culpa a la ia https://t.co/kp2c60gX7e Pasad vuestros scores, a ver quién saca un 100 👀

NERVE NOW READS EVERY TWEET BEFORE YOU DO. I ADDED A CT NARRATIVE ANALYZER AND IT CAUGHT A META SHIFT 47 MINUTES BEFORE THE FIRST VIRAL THREAD. +0.9 ETH. the pipeline was fast. jev scores in 190ms. sentinel kills honeypots for free. grok reads telegram and discord. but ct moves faster than all of them. a kol tweets. quote tweets stack. the meta shifts. by the time it hits telegram the entry is gon

Codexリセットこないにゃ(-ω-；)ｱﾚ? …まぁいいか🤣 昨日からずーーっと調整してたAIによるナレッジベース自動整理が完成にゃ😸🎉 毎月AIが自動起動→情報整理→Jevで分類→自動処理。僕の判断が必要な時だけ🔔へ 文字だけじゃつまらんので、パックマンみたいにゴミを食べて整理するUIも実装にゃ😹 正常なら100%で静かに終了✨