The Model That Can't 'Chat' While everyone chases long reasoning, TypeSafe AI has emerged from stealth

AI Release 🤖 The Model That Can't 'Chat' While everyone chases long reasoning, TypeSafe AI has emerged from stealth with Jev — the first 'System One' model: it doesn't generate text at all. Only typed decisions with probabilities. How it works. You feed it a state (logs, a ticket, JSON) plus questions of three types: Choice — pick from a list, Score — rate on a scale, Noul — yes/no as a probability. An answer in 70–500 ms, priced at $0.042 per million input tokens, output is free. Everything is resolved in a single parallel pass, with no text generation. Why this matters. Most decisions in code are 'System 1' judgments ('which bucket?', 'is this urgent?'), and until now you'd rent an expensive 'System 2' — an LLM that writes text for you to parse. Jev returns a code-ready value plus calibrated confidence. It was trained with RLCD: if the model claims '90% confident,' it's right about 90% of the time. And yes, it 'can't hallucinate' — it won't return a value outside the given schema (though it can still make a wrong choice). Who's behind it. Diogo Almeida — co-author of InstructGPT, the method that turned ChatGPT into an assistant. $40M seed from DCVC. Access is still whitelist-only, but it's already available via Vercel AI Gateway. And the main trend. Within days, dozens of repositories have sprung up around Jev on GitHub: Claude Code plugins, CLIs, awesome lists, use cases. The ecosystem of 'machine-native' models is just beginning — and this may be the next big branch of AI. 🔗 Introduction to System One Models and Jev #AI #Jev #SystemOne #neuralnetworks typesafe.ai Introducing System One Models & Jev - TypeSafe AI Blog TypeSafe AI is an AI lab building machine-native intelligence infrastructure for automation, designed to make decisions within software. Try our first System One Model, Jev, in early access. 12 views 06:17

🔗 Read on t.me