GPT-6 Astra reportedly learns robot tasks from video without a text prompt

A user describes GPT-6 Astra performing in-context learning for mobile manipulation: inferring what to do from video without a text prompt. They say it handles a different environment, camera angle and layout, and chooses between directing the robot’s working end and controlling its joints. Another user sharing the demo calls the development “amazing for deployment speed” and “frustrating for foundational models.”