Hackaday · Creativity & design

Jev: A New Breed of Fast, Cheap Decision-Making LLMs

A new category of AI models, exemplified by the recently released Jev, is emerging. These models are specifically designed for rapid and cost-effective decision-making and classification tasks, rather than generating text.

Unlike traditional LLMs that produce language, these new models accept text input but output only floating-point numbers. These numbers represent answers to user-defined questions, such as yes/no queries, multiple-choice selections, or scoring requests.

Jev is noted for its extreme speed and low cost, focusing solely on its decision-making capabilities. It cannot generate text.

The concept has quickly gained traction, with developers creating their own decision-type models and conducting experiments. Examples include Kev and Nimble, with Nimble recently added to Ollama and capable of running locally.

This development suggests a potential new avenue for AI applications requiring quick classification and decision processes.

AI-samenvatting op basis van de bron.