Jev is the FIRST of a Whole New Class of AI Models (Here's How to Actually Use It)

YouTube · Watch/Read · (gathered 2026-09-23)

• Cole introduces Jev as the “FIRST of a Whole New Class of AI Models” called system one models, which are master decision-makers, distinct from Large Language Models (LLMs). He emphasizes its genuine innovation, speed, cost-effectiveness, and reliability for decision-making, addressing common LLM issues like hallucination in this specific context.
• Jev operates by taking a situation and multiple-choice questions as input, then outputs a decision with a confidence score, using a new RLCD (reinforcement learning for calibrated decisions) training algorithm. This makes it significantly faster (20-200x), cheaper (40-1000x), and more reliable (0% structured output failure rate) for decision-making compared to LLMs.
• Practical use cases include customer support routing, playing video games (demonstrated with Cole’s own game, Doom, and Pong), browser automation, and acting as an LLM router to optimize costs and speed by directing queries to the appropriate LLM. Cole integrates Jev into his AI coding workflows for tasks like PR triage and game testing within his Arkon framework.
• Jev is not a replacement for LLMs as it cannot generate text or create content. Instead, it’s best used in combination with LLMs, where Jev handles decision points and classification steps, while LLMs manage other reasoning tasks, creating more efficient and reliable AI workflows. Its “general classification” capability distinguishes it from traditional, task-specific classification models.