Elorian 发布首个模型 Glimpse，面向视觉思考

Elorian’s first model, Glimpse, is built for Visual Thinking. Ask it questions about images and video, and it reasons through what it sees. How many monkeys can you count? The part I’m interested in: being able to dig into footage. What changed? Why did that happen? What is the robot doing at each step? If it can do all that reliably, robotics data labeling changes entirely. Most teleop footage sits unlabelled because segmenting every step by hand is slow and expensive. A model that watches the clip and writes the labels itself turns that backlog into training data. Feels like there’s a lot to build here. cc @ElorianAI @AndrewDai

来源：X · Linus Ekenstam · x.com