Jev Drives a Simulator in Real Time
The author connected Jev to the raw controls of a driving simulator and shows it driving in real time, emphasizing real-time response rather than frame-by-frame video generation.
Jev focuses on type-safe decisions, potentially boosting multi-sensor autonomous driving
The post notes that Jev drops text generation and is designed for fast in-software decisions: unstructured input, type-safe structured output with calibrated probabilities, and asks whether it could judge next actions from multiple autonomous-driving sensors.
Jev 完全放弃了文本生成能力，专门为软件内部的快速决策而设计：输入非结构化数据，输出带有校准概率的类型安全结构化结果. 会不会发展出一种模型对自动驾驶的不同sensor的输入做出下一步行动的判断，从而使lidar + mmWave radar+camera这一条自动驾驶的路线成为主流？
· 0 likesOpen on X
More like this
Using TypeSafe Jev to make a Rubik's Cube solve itself like a human
A developer used TypeSafe Jev to make a Rubik's Cube solve itself with a beginner method in 94 moves, like a person rather than the 22-move optimal solution; Jev is not an LLM and answers one question with a probability in ~250ms.
Controlling a LeRobot Arm with JEV Decision-Making
A developer experiments with GPT for reasoning and JEV for decision-making to control a LeRobot arm, with updates promised soon.
