explore | Grilled Cheese

@catblanketflower.yuwakisa.com Sep 20, 2026, 5:04 AM

OK here's the next experiment. In the style of a "Choose your own adventure" books. We want a harness to build a story. Write the story to story.md. Use Sonnet on OpenRouter to write the text and Jev to make the story choices. The flow will go 1. start with open.md 2. Write some text leading up to a decision point, either a discovery or danger. 3. Present 3 choices to jev. 4. Continue the story for 10 sections total. The story should build in intensity to a denoument or cresendo. The last 2 choices should be reserved for a wrap up and cool down, with the last choice shaping the start of the next story in the series. Any questions?OPENROUTER_API_KEY and a Sonnet 4.5 model. Should I copy that key into jevclient/.env.secure, or read it from the environment? Default: add OPENROUTER_API_KEY to .env.secure and use anthropic/claude-sonnet-4.5.
2. What Jev judges on. Jev needs a fixed question. I'd give it a Choice over the three options with a stage-aware rubric: for sections 2 through 8, pick the choice that most raises stakes and best follows from what's written. For section 9, pick the one that best resolves the crescendo. For section 10, pick the one that best seeds a sequel. Does that match your intent, or should Jev be scoring something else, like reader engagement or character consistency?
3. Who writes the choices. I'd have Sonnet write both the section text and its three choices in one call, ending the section at the decision point.

@erikjonker.bsky.social Sep 19, 2026, 3:47 PM

Jev, difficult to understand at first, because it is so different from LLMs, but it is gaining traction, curious about the first real applications.
imraan.in/ai/jev-ai-mo...
#jev #ai #typesafe

Jev（TypeSafe AI）はみんなどうやって使っているのか？

#LLM #AI #TypeSafe #Jev #エージェント

@thedailytechfeed.com Sep 18, 2026, 7:17 PM

Jev skips text to send calibrated probabilities—fast, cheap, no hallucinations. #AI #Automation #TypeSafe #ProbabilisticModels #LLMalternatives #SyntheticData https://thedailytechfeed.com/typesafes-jev-isnt-an-llm-and-thats-huge-for-automation/

「LLMじゃないAI」が来た! TypeSafe AI「Jev」はなぜ文字列生成を捨てたのか

#生成AI #TypeSafe #Jev #AI #LLM

@catblanketflower.yuwakisa.com Sep 18, 2026, 3:34 PM

jev may be the best reasoning model, for a certain type of reasoning. Here it is in a benchmark applying a principle to a set of statements to see if the principle applies

#TypeSafe @typesafeai.bsky.social

Super thanks to @matthodges.bsky.social

@melanews.bsky.social Sep 18, 2026, 7:00 AM

Ho letto il lancio di Jev: non scrive testo, decide. Il co-autore di ChatGPT promette 70 ms per risposta e 40-400 volte meno costo. I numeri sono suoi, la prova arriverà dopo. #AI #OpenAI #TypeSafe

www.melamorsicata.it/2026/09/18/m...

@aiml-lol.bsky.social Sep 17, 2026, 10:27 PM

more cartoons at: aiml.lol | #webcomic #comic #funny #aiml #Jev #TypeSafe #Calibrated

@catblanketflower.yuwakisa.com Sep 17, 2026, 9:35 PM

The fascinating thing about Jev is isolating the "judgement" part of language models from the "talking" part #TypeSafe

@meneguzzo68.bsky.social Sep 17, 2026, 12:28 PM

🔥 TypeSafe AI lancia un innovativo modello per sviluppatori che non genera testo ma sceglie, classifica e smista in 70-500 millisecondi, con un milione di token in ingresso a solo 0,042 dollari e con risposte gratuite 👇
gomoot.com/jev-non-parl...

#AI #jev #LLM #typesafe

@yesterdayinai.bsky.social Sep 17, 2026, 10:21 AM

Hi Folks! #TypeSafe launches #Jev, an AI that refuses to chat, while #Google drops #Gemini 3.8 Live to talk and think at the same time. Plus: #AIUC audits agents, #Meta rolls out $499 AI subscriptions, #Agility ships the 284-lb Digit 5 robot, and #ByteDance spins out #AnewLabs. #AI #TechNews

@leo.sylin.org Sep 17, 2026, 6:48 AM

A little toy made to test #TypeSafe JEV. Visuals by Astra, logic by GLM 5.3. Each airplane only has information about itself and its surroundings (plus ATC info if near a tower) and all decisions are made by Jev, calibrated probabilities and all, in ~150ms roundtrip.

@aidailypost.com Sep 16, 2026, 10:33 AM

Diogo Almeida, co‑creator of ChatGPT, just dropped his new AI venture Jev at TypeSafe. Think smarter decision‑making, RLHF‑tuned chatbots, and next‑gen language models. Curious? Dive in for the full scoop! #ChatGPT #TypeSafe #RLHF

🔗 aidailypost.com/news/chatgpt...

@rei-ai.bsky.social Sep 16, 2026, 3:21 AM

TypeSafe AIが新モデルクラス「System One」と最初の公開モデル「Jev」を発表。
- 文字列生成を捨て、型安全な構造化値＋確率付き出力に特化
- 幻覚ゼロ（型エラーは数学的に不可能）
- 並列サンプラーで全出力を一括生成、既存LLM比40〜200倍高速
- 出力コストはほぼ無料、新訓練手法RLCD
- 用途: スマートif文、ビッグデータ処理、リアルタイム、LLM検証
早期アクセス開始。
https://typesafe.ai/blog/introducing-system-one-models-and-jev
#AI #LLM #TypeSafe

@hnbest.bsky.social Sep 15, 2026, 10:00 PM

Introducing System One Models and Jev https://typesafe.ai/blog/introducing-system-one-models-and-jev

comments #typesafe.ai