Resources for building with Jev

Diogo Almeida

@CompleteSkeptic

Types Beyond Decisions: box We’re also releasing a new type that is somewhere in the gray area of decisions: a single bounding box. We warned you there’s a reason we call them System One Models! (only on jev-preview)

Alex Warren

@exrhizo

New @typesafeai Jev capabilities! Some decision models can ask questions about images but can they ask questions with images? 🧵 How to get the most out of the new Jev launch

Nathan LeClaire

@dotpem

Here's another JEV TIP in honor of JEVDAY Jev can answer as many questions as you can fit into context in parallel. You can ask hundreds in one call, and since output is free, why not? 100 questions on a 32k token input document: ~$0.00001 per question ~1ms per question

Matt Canham

@matthewcanham

I've had many people reach out saying they've read everything they can about Jev but still don't get it. Here's an explanation for the normies

Avid

@Av1dlive

everything you need to start building with jev, in one article. code, architecture, diagrams... everything you need to follow the build and make it your own.

GREG ISENBERG

@gregisenberg

Jev is HERE and this is the CLEAREST explanation of what it is and what NEW businesses it unlocks. (and at the end I'll tell you how to get Jev even if you're on the waitlist) WHAT IT IS You know how you open your inbox and have to decide what's junk, what needs a reply, and what can wait? Jev does that part. It looks at each thing and says "this is junk, I'm 94% sure." It doesn't write anything back to you. It just sorts. 1,700 emails for 18 cents, instantly. That sounds kinda trivial but the important part WHAT IT UNLOCKS My explanation of Jev sounds small until you realize HOW MANY jobs are exactly this. Someone reading a stack of applications. Someone deciding which support ticket goes to which team. Someone looking at inbound and deciding who's worth calling back. A few ideas on what it unlocks: 1/ Instant quotes that are actually instant. Every quote form on the internet says "we'll email you by end of day." Build the version that answers in under a second, for roofers, movers, insurance, legal intake. 2/ Lead scoring as a product. Every agency and service business has a contact form full of junk. Score every submission and send the real ones straight to the owner's phone. 3/ Support triage for companies with no support team. The ticket gets classified and routed before anyone opens it. 4/ Clipping tools. Pass in a transcript, get the best moments scored in three seconds. Every clipping product just got a cheaper engine. 5/ Application piles. Grants, permits, insurance claims, job apps, loan docs. Someone reads that stack one item at a time today. 6/ Marketplace matching. Someone types what they need and gets matched to the right local business instantly instead of waiting for callbacks. 7/ Browser agents that actually move FAST. That makes bulk browser work practical: pulling quotes from five carriers, filing the same form for 200 clients, checking supplier inventory in real time etc. TLDR; find an expensive queue and put Jev at the front of it. HOW TO GET IT I didn't realize you can skip the waitlist because Jev is live on the Vercel AI Gateway right now, so you can start calling it today. In this episode, we share how. Episode now live on @startupideaspod (thanks to @ryanvogel for coming on and spilling the sauce today) Watch: https://www.youtube.com/watch?v=4mTLpuQpB80 Jev is a big deal because this is a whole new way to do AI Really cool Happy Jev day.

iwashi / Yoshimasa Iwase

@iwashi86

Jevの内部アーキテクチャを推測している技術記事(Jev’s Architecture Unmasked)からメモ。 ・本記事はJevのAPIを約1万回の呼び出して、内部構造を推測したもの ・従来の言語モデルを用いた分類やルーティングでは、トークンを1文字ずつ逐次生成するために膨大な無駄な計算コストが発生していた。 ・JevはLLMの膨大な知識をそのまま保持しつつ、テキスト生成を完全に排除して内部表現から直接確率を出力する ・システム全体の構造は、共通の文脈をエンコードする共有状態、並列に処理される質問群、そして確率を直接読み出すヘッドの3要素から成立 ・モデルは入力を左から右へ読む事前学習済みの因果トランスフォーマーであり、プロンプト処理の直後に推論を終了する ・テキスト生成ループは存在しない ・だから、単語の綴りやJSONのフォーマットを出力するためのステップが一切不要 ・APIレスポンスに含まれる出力トークン数という項目は、単に返却された文字列の長さから、後で計算された課金用の指標にすぎない ・選択肢が2個の場合でも200個の場合でもサーバーの処理時間はほぼ変化せず、モデルがトークンを逐次生成していない事実を強く裏付けている ・入力データは共有状態と複数の質問に分割され、問い合わせ本文などの共通コンテキストは一度だけKVキャッシュに保存される ・複数の質問は同じ共有キャッシュを参照するため、質問数が増加しても共有状態を何度も再計算する必要がない ・それぞれの質問はお互いの存在や指示が見えないようになっている ・質問同士が干渉することなく完全に隔離された状態で推論される ・ある質問内に秘密のコードを記述しても別の質問からは参照できない ・共有状態に置いた場合のみ全質問から認識可能 ・約100個の質問まではリクエストの処理時間がほとんど増加せず、共通状態を一度だけ計算して並列処理する設計の恩恵が明確に現れている ・Jevが持つ高度な知識水準を考慮すると、双方向エンコーダを新規開発したのではなく、既存の因果デコーダ型LLMを再利用した可能性が極めて高い ・トークナイザの挙動は既存の公開モデルと完全には一致しない ・数字の分割方法やチャンク処理に独自の工夫が見られる ・選択肢の並び順に関する実験では、判断の根拠となる情報が選択肢リストの末尾に配置されていてもモデルは正しく正解を選択できた ・質問に含まれる各選択肢は独立して採点されるのではなく、リスト全体が互いに文脈として影響を及ぼし合った上で判断される ・無関係なダミーの選択肢を追加すると既存の選択肢間の相対的なオッズ比が変化した ・このことから、選択肢同士が相互作用していることが実験的にわかる ・リスト全体を一括して読み込む設計のおかげで、上記のどれでもないといった他の選択肢に依存する回答も適切に処理できる ・選択肢の提示順序を逆転させると出力確率に若干の変動が生じるため、実務で特定の閾値を用いて自動処理を行う際は順序バイアスへの配慮が求められる ・学習手法にはRLCDと呼ばれる独自のアプローチが採用されている ・強化学習の枠組みを用いて真の結果に即した予測確率を出力するように調整されている ・モデルの出力は対数損失やブライア損失といった適格スコアリングルールで最適化され、正直な不確実性を反映した確率分布を形成する ・ベンチマークの検証結果において予測確率と実際の正解率は高精度に一致しており、期待キャリブレーション誤差は0.0313という極めて低い値を記録した ・難易度の高い数学問題ではモデルが提示する正解確率自体が低下する ・なので、自身の知識の限界に応じた不確実性を表現できている ・APIが返す信頼度スコアは学習された固有値ではなく、得られた確率分布の最大値から算術計算で求めた便宜的な数値にすぎない ・Jevのバックボーンには、必要な専門モジュールのみを選択的に起動するスパースなMixture of Expertsが採用されていると推測される ・テキスト生成を行わないプロンプト処理のみのタスクでは計算負荷がボトルネックとなるため、疎な構造による演算量削減が大きな効果を発揮する ・約3万トークンの入力をわずか160ミリ秒程度で処理する速度性能は、巨大な密結合モデルではなくMoEを採用している推測ともあう ・質問枝はチャットのような逐次的会話ではなく独立したバッチとして処理されるため、GPUリソースを最大限に活用した一括計算が可能 https://archerhume.com/posts/jevs-architecture-unmasked

Moritz Kremb

@moritzkremb

Full Jev Tutorial What it is, how you can build with it and what new applications it can unlock → 0:00 Intro → 0:34 Jev explained → 4:06 API setup → 5:59 Demo 1: Voice-controlled browser → 11:33 Demo 2: AI memory → 17:27 Demo 3: YouTube predictor

cocktail peanut

@cocktailpeanut

Jev is cool not because it re-invented classification, but because it makes ARBITRARY classification into a type-safe programmable primitive. A general purpose zero shot decision model whose native interface is RUNTIME-DEFINED typed decisions, optimized for that exact interface

Paarangat

@paarangatrai

this is the easiest way to understand Jev: LLMs generate answers. Jev makes decisions. that sounds like a small difference, but it actually changes the entire use case. say you give a normal LLM this: “here’s a user, their account history, payment behavior, support chats, device data, etc. tell me if this looks risky.” the LLM might reason through it and return: “yes, this looks high risk.” maybe in JSON if you ask nicely. with Jev, you define the possible decisions upfront: risk: * low * medium * high manual review: * yes * no and Jev returns something closer to: risk = high (96%) manual review = yes (91%) that’s basically the product. it’s not trying to be another ChatGPT. it’s more like an AI-native if statement. instead of: if transaction > $10,000: review() you can start thinking more like: if “does this behavior look suspicious?” > 95%: review() and that opens up a pretty interesting category of software. a few assumptions I had at first that turned out to be wrong: 1. “so it’s just a classifier?” kind of, but that undersells it. the input can be messy real-world context, and you can ask multiple typed questions about that state at once. fraud? churn? escalate? eligible? priority? all from the same input. 2. “so it replaces GPT / Claude?” not really. I actually think the interesting architecture is: Jev decides WHAT needs to happen Claude / GPT reason or generate WHEN deeper intelligence is needed normal code executes the deterministic stuff. Jev becomes the routing layer. 3. “it can’t hallucinate?” this one needs nuance. if your allowed answers are: LOW MEDIUM HIGH Jev won’t suddenly invent: “EXTREMELY HIGH 🚨” the output structure is constrained. but it can still be wrong. HIGH at 92% can still be the wrong decision. so “no hallucinations” doesn’t mean “always correct.” 4. “why not just force an LLM to return JSON?” you can. we already do this everywhere. but you still deal with generation latency, schema validation, retries, weird outputs, confidence estimation and a lot of glue code. Jev is designed around the decision itself rather than text generation. 5. “why should I care?” because most software is ultimately a giant tree of: if this → do that if this → route here if this → escalate if this → reject if this → ask a human Jev is basically asking: what if those if statements could understand messy human context? that’s a much more interesting framing than “another AI model.” I can see this being very useful for: fraud / risk support routing moderation PR / QA automation lead scoring compliance workflow orchestration agent routing especially as the cheap + fast decision layer sitting in front of larger reasoning models. early tech, obviously. but the category itself makes a lot of sense.