Jev for search — a daily feed

I built something fun using JEV I fetched the Telegram Channels in my community. You type what you are looking for. A topic, a question, or just a vibe. The channels that fit the question rise up in seconds. https://t.co/gjEdDkIsYh

（続）Jev は、RAGとGraphRAGをどう変えるのか？ もうちょっとだけ詳細に調べてまとめてみました。 1. リランカーとしては強い NanoHotpotQA（50問、候補100件）で、関連の薄い文書を92%除去しても正解97件はすべて保持。nDCG@10は0.833から0.975に上がりました。 2. ただし、候補にない文書は救えない 上位8件だけをJevに渡す方式では、45位にあった正解を一度も見られず、答えられませんでした。判定を磨く前に、検索の網羅性を確かめる必要があります。 3. GraphRAGでは「嘘の関係」を切れる 否定文から抽出された関係を判定して除去すると、Precisionは0.279から0.404に。Recallは変わらず、落としたのはノイズだけでした。 4. 名寄せの本当の問題は、候補づくりにあった 判定は完璧でも数値は動きませんでした。型が違うため、比較す

TREC, DBPedia & Banking77 × 4 target LLMs × 1–4 shots: ✅ best or tied-best in 44/48 settings ✅ beats its own Semantic TopK start in 46/48 📈 avg 1-shot acc 88.1 vs 81.4 (CASE) and 81.2 (Semantic TopK) Biggest gains when the demo budget is tightest. Bar chart of average 1-shot accuracy on TREC, DBPedia and Banking77 across 4 target LLMs: Jev-LDE 88.1, CASE 81.4, Semantic TopK 81.2, TopK+ConE 78.3, L

6 search strategies competing to find hidden "reward pockets" in a randomly generated maze, with fixed budget of tries. A 60-year-old math (MCTS) crushed everyone. Qwen) did okay. Why poor new JEV kept repeating the same choice and lost even to pure random guessing. @typesafeai https://t.co/LwahJrDKV6

@SubhadipShips Just completed Jev for both Text and Images. Ranked #1 model on JevBench and #3 on DecisionBench. https://t.co/pSmvDp1X4F

Agen web jev-ultrafast raih 21.276 star dalam 13 hari. Tolok ukur pencarian tiket maskapai tuntas dalam 7 detik dan 3 langkah. Sistem pangkas DOM 3 lapis potong 94 persen konsumsi token ke model. https://t.co/uMGOrmv05q

RSI-Jev v3.0 is out, and this is the first version where RL actually helped. Our AutoScientists loop ran 59 RL experiments on the 2B decision model. Most failed. The pattern was pretty clear: when every example already gives the model the right answer, plain supervised training is very hard to beat. What finally worked was giving RL a signal the labels didn’t contain: whether the model ranked the

Do you 🫵 get lost in 10 concurrent Codex threads? Using Jev from @typesafeai, I can jump instantly to an old thread when I forget it. Open source! https://t.co/faoLlPw641 https://t.co/b1HskBiFDg

随便测试了一下，对比 amp 的代码搜索子代理，优势是不遗漏，但是搜索结果有点多；而且很贵啊，跑了 3 个问题花了我 1.4 刀，说好的 jev 很便宜呢。。。 https://t.co/c3MSwnwmDg

My X feed has been full of Jev as an embedder, so I got curious. Can good decisions make good coordinates for search? Turns out, they don't. That said, Jev is an outstanding reranker and post-filter. https://t.co/XtuF1FvGxU

Opusに感想文かいてもらいました AI エージェントの長期記憶の検索を判定モデル Jev に任せ、入力トークンを 1/17 に削るまで｜こかげ https://t.co/4NgSsKpDZO #zenn

I spent a week building world’s best FREE indexer of successful AppStore apps, with Jev as the driving force behind its speed: https://t.co/fWHIEwuRpz

最近話題のJevを使って、アプリを作ってみました。 好きなテーマで、いろいろランキングしてくれます。 よかったら遊んでみてください🙌 https://t.co/hsq4bCBjL3

https://t.co/i9BX1Bp58b https://t.co/627YaFep3n é bem simples, fiz um app pra analisar algumas fontes que gosto e avaliar quais artigos são realmente relevantes pra mim uso o jev para classificar cada artigo em features/importância e depois para detectar se dois links cobrem a mesma cois

@_avichawla I've developed a Jev for agentic rag in financial research, results: 50/50 right, 100x fewer tokens, up to 18x cheaper than an agent. Tested on 4 companies and 4 LLMs, including one running locally. https://t.co/L0rXy3WVgX

Çok yeni pişirdim! Haber trade yapanlar gerçekten iyi para kazanabiliyor. Özel yazılım yaptığım bir müşterim “tek bir haberle emekli olacağım” derdi. Şimdi AI haberleri analiz edebilecek kadar hızlı ve akıllı. JEV AI ile yaptığım NewsWithJEV uygulamasını ücretsiz kulüp üyeleriyle paylaşacağım. Skool: https://t.co/AZw925QLgz

He metido Jev en un buscador de emojis para poder escribir "lunes otra vez" en vez de buscar "cara triste" Completamente overkill pero me encanta como funciona https://t.co/mm7EpPOyWZ

複数キーワードをJevに投げれば、キーワードごとに関連度をグラフで表示できる。 キーワードの類義語もサポート可能。 棒グラフからどんな公報か理解できるため、読むべき公報、スルーできる公報が素早く判断できる。 ←左：複数搬送車無線通信制御の公報 右→：構造物の２次元地図作成システムの公報 https://t.co/VQkwCutir2

https://t.co/PWJnXLHONg 1. upload your resume 2. get an Opus 5.5 sizzle reel of your work experience search profiles powered by Jev free while i can afford it https://t.co/mwkBZMqsbo

4/ 重排，我测的： 20 道有答案的题，排第一的是答案：BM25 7 道，Jev 重排后 17 道。前 3 也是 17 道。 17 就是上限，另 3 道 BM25 前 10 段里根本没有答案。以 0.5 为线过滤，10 段平均只留 2 段给大模型。 注意 BM25 偏弱，换好的向量检索差距会变小。 https://t.co/LAHTaaceWt

Made jev-rankkit for reranking text. Then realized: not everything is text. So I built TypedRank - rank tool calls, tickets, SQL schemas, anything. Runs on Jev + Laya, works standalone too. GitHub: https://t.co/6jWPfO1409 #BuildInPublic #OpenSource #AI

Right now everyone's wiring @typesafeai's Jev into chatbots. I used it to replace Ctrl+F: ask any page in plain words → the exact sentence gets highlighted. 107 passages, exactly 1 match, 1.5 seconds. Every answer quoted from the page. Open source ↓ https://t.co/vLip3lAPXw

Introducing ProbablyThisWay 🏔️ built with @typesafeai 's Jev Model. > you can search the trails in any area on earth > find the best trail according to the Current state: temperature, wind, and every candidate route’s metadata like distance, estimated time, etc using Jev https://t.co/D6a0wh21tL

I expected a better reranker to make RAG-Fusion redundant. So I tested Jev as the reranker on 200 NFCorpus queries, and got the opposite: fusion's lift doubled (+0.050 NDCG@10, against +0.025 with bge-large). https://t.co/QiGPhWZLJj https://t.co/QdfVcDvNbo

We’re finally live! Check out this free Jev-powered AppStore indexing tool: https://t.co/wW8vdTrDWC

Added Jev in the development loop for slicing stories and prioritizing https://t.co/7BFeF0buHu

* LLM概要生成なし。Jevランキング記事 追加しました。 https://t.co/LWo6YlpkDx

Go Proposal Weekly Digest に Jev 使って検索機能つけたので、その解説記事です。 https://t.co/QXUgchOQtR

Added an open search history to the https://t.co/6rvbCXTB3W site so everyone can see what others are searching with Jev. Pretty fun to see who's searching "what" from "where". https://t.co/QOl9Jqbz2x

jev + opus 5.5 on stream clips is insane still can't believe every streamer isn't running this already on a 3-hour vod it turned an evening of scrubbing into about 5 minutes, and the bill came out under a dollar. easily the best use i've found for it so far → jev looks at each frame and just says if something's actually happening → if it's a real moment, the clip gets cut right there → long boring

Jevを活用した「プレ知財分析アプリ」を作ってみました。 使い方はシンプルに。 1. ざっくりとした母集団データをアップロード 2. 知りたいことを自然言語で入力 （例：走行安定性に関する制御） 3.Jevが情報を整理して再分析 700件のJev再分類で$0.0457 7円程の額でした！ ※5＄で100回強 本格的な知財分析の前に分析イメージができます。 知らない技術の分析時にFIやキーワードのアタリを付けることにも役立つかもしれませんね。 #知財分析 #Jev #Opus55

Visual search with Jev. $0.00023/search -> $0.23 for 1,000 searches -> 10m searches for 1m user ~= $2300 total 🤔

We just fixed product research with JEV I was tired of seeing JEV slop so I built something useful Tech part JEV + Shopify UCP + Meta ads lib RT + comment "JEVSLOP" and I'll send it to you rn https://t.co/ke11tLbufi

TypeSafe Jevとソーシャルシグナル（Hugging Face）による週次arXiv多段階ファンネル選別の実践とアーキテクチャ https://t.co/jHYuWlrjd6 Jev楽しいねぇ。Jev-as-a-JudgeをarXivの週次キャッチアップのワークフローに組み込んでみた。

Jev changed how I keep up with AI news. I used it to build a tool that scans every new AI post on X and quickly surfaces the ones taking off across the industry. So you can spot what’s breaking through and join the conversation early. Repo : https://t.co/Lzvgmh7ql2 https://t.co/VXLWcCnTka

Must Watch - https://t.co/9lCMFx5I2C Lots of things I talked about and complete end to end build I Built an AI Agent That Does Market Research — Jev + LangGraph + https://t.co/17JP6wMCkc https://t.co/DRxpHgfEz5