CompanyJEVJev

TypeSafe AI: Jev makes judgments, code makes decisions

Original title: Don’t Call It a Decision Model

Plain language

What it is: TypeSafe AI explained why its Jev model is called a System One model. The company prefers this term over decision model because Jev is built to make quick, intuitive judgments, while computer code makes the final decisions.
How it works: Instead of asking the model to make a complex decision all at once, users break the decision down. Jev answers simple questions that require human-like judgment, and traditional code handles the exact mathematical or logical rules.
Where it is available: This approach applies to software systems and workflows integrated with the Jev model, which is accessed through the TypeSafe AI platform and API.
Example: In a customer service system, Jev judges if a customer is asking for a refund, while the code checks if the refund amount is under twenty-five dollars to decide whether to grant an instant refund.

What it is: TypeSafe AI defined Jev as a System One model designed for atomic judgments rather than end-to-end decision making. This design avoids complex prompt engineering by separating cognitive judgment from deterministic business logic.
How it works: Developers decompose a decision path into isolated Jev queries for subjective questions and standard code for logical conditions. The system combines these outputs, which improves observability through detailed execution traces showing each judgment probability and code fact.
Where it is available: This architecture is used when building applications with the Jev model, specifically using the jev-latest model via the TypeSafe API.
Example: A workflow routes a user to a fraud team by combining a Jev judgment on whether the user reported an unauthorized charge with a code check on whether the refund is eligible.

The headline and explanation on this page are based on TypeSafe AI's post. The full details are in the original.

Don’t Call It a Decision Model