Shut up and calculate: Jev's new AI primitives for coders

TypeSafe's Jev, launched last week, differs from standard frontier models and has attracted scrutiny from developers exploring its capabilities. Jev is a classifier with brains, returning predefined decisions, probability distributions, and confidence levels based on the input data. Developers find Jev's structure appealing, as it requires manual configuration and results in fast, highly structured decisions.

Jev operates as a "System One" model and can respond to three types of questions: Choice, Score, or Noul. The structured output from Jev offers advantages such as faster response times and lower costs compared to standard LLMs. Potential applications include job recruiting, scientific paper screening, and managing software that would otherwise require human intervention.

While Jev is fast and cheap, its accuracy remains uncertain, and it may be best suited for static, self-contained training sets, such as spam detection or online shopping recommendations.