JEV-27B-VL 双系统：快决策 + 深推理
AutoTrustAI 解释为何用 Blocks of Experts 把 System 1 的校准式 yes/no、多选、0–5 打分与 System 2 的 Qwen 视觉推理放进同一引擎。
Why build this with Blocks of Experts? JEV-27B-VL keeps two complementary capabilities in one engine: System 1 Fast, calibrated yes/no decisions, 2–16-way choices, and 0–5 scoring over text and images. System 2 The full Qwen3.8-27B visual reasoning path for tasks that need deeper analysis, generation, or step-by-step thinking. Fast decisions should not replace deeper reasoning. A capable agent needs both — and needs to know when to use each.
A decision model matters when the environment keeps changing. In this Tetris demo, JEV-27B-VL reads the board, evaluates candidate placements and orientations, assigns probabilities to each action, and chooses the next move as the game accelerates. Result: 20/20 lines cleared with 57 pieces. This is not a chat demo where an AI explains the board after the fact. It is a visual decision loop running continuously while the task is in motion.
Introducing JEV-27B-VL. We believe it is the world’s first open-weight, near-SOTA multimodal decision model — built with AutoTrust AI’s Blocks of Experts (BoE) recipe. It does not just describe what it sees. It turns visual state into calibrated action probabilities, then selects what to do next. In this demo, it sees Mario’s position, movement, obstacles, and timing — then chooses whether to run, jump, or do both. See → decide → act.