TypeSafe AI Argues Reliable Decisions Require Code, Not 'Decision Models'

Key Info

TypeSafe AI published a thread explaining why it does not call its product Jev a "decision model," arguing that reliable decisions must be built by composing reliable judgments in code rather than delegating decisions to a model.

Highlights

• The team says treating Jev's questions as decisions uses the wrong layer of abstraction and hurts reliability.
• It argues semantic logic hidden in string caveats, embedding vectors, or complex ML cannot outmatch real-world data distributions.
• Its recommended approach: be clear about what you need to know and confirm a plain reading of the data is enough.
• The thread frames this as the basis of its "System One" approach to reliable decisions.
• A follow-up post reiterates that Jev is not a decision model.