Author(s): Enzo Lombardi

Originally published on Towards AI.

The classifier that never writes a word

TypeSafe’s Jev announcement did the rounds properly: a frontier model that does not generate text, returns typed values with calibrated probabilities, and answers in 70 to 500 milliseconds instead of the 3 to 329 seconds a generative model spends thinking out loud. The pitch is a good one. Unstructured state goes in, a typed decision comes out, and there is no JSON to repair on the way back.

The author argues that classification-style “System One” models were possible long before the recent hype, and that the core advantage is structural type safety: the model scores options via marker positions rather than generating tokens. They explain how Laya works, why prompting should be verbal sentences (not raw JSON or labels), and how option phrasing and calibration affect accuracy and thresholds. The post also covers implementation details in a pure-Rust crate, performance costs driven mostly by context/state length (not number of questions), and two practical demos: credential-exfiltration triage and a real-time Pong agent. Finally, the author emphasizes design rules—don’t ask the model to infer unseen policy, test wiring with known tasks, and view it as a typed decision system with probabilities you can threshold—while noting limits like adversarially crafted inputs.

Read the full blog for free on Medium.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI