Generalist's robot can learn tasks on the spot—without training

generalist ai, a startup founded by former google deepmind and boston dynamics engineers, demonstrated robotic arms capable of performing tasks immediately after viewing a short instructional video. the robots showed significant adaptability, improvising by using a banana as a tool or switching grip angles when a standard approach failed, bypassing the thousands of training examples typically required for robotic manipulation. the company collects this data using specialized camera-equipped gloves worn by human workers to capture high-quality physical interactions, aiming to create a general model that understands world physics like a human. this development signals a potential move away from brittle, task-specific training toward flexible, generalist robot intelligence capable of operating in varied, real-world environments.