# Robots That Can Learn on the Spot Generalist AI's robots learn and adapt quickly, just like humans. - **Source:** [Wired](https://www.wired.com) - **Published:** 2026-08-19 - **Topic:** [Data & AI](https://brifing.si/c/data-ai) - **Original article:** - **On Brifing:** https://brifing.si/s/dafca883-939d-4b3b-9159-23d42c43c092 ## Robots Learning Like Humans Generalist AI's robotic arms can quickly learn simple tasks, much like humans. For example, they can watch a short instructional video and figure out how to stack cups or place blocks into bowls on their own. One of the most striking examples was a robot using a dustpan to sweep a block into a bowl when the brush was removed. This demonstrates how robots can adapt to unexpected situations. ## Approach Inspired by GPT-3 Generalist AI's CEO Pete Florence compares the robots' learning ability to OpenAI's GPT-3 model. Just as GPT-3 can instantly perform new tasks, these robots can quickly grasp new challenges. The company focuses on teaching robots the physical rules of the world, similar to the intuitive understanding of physics humans develop from childhood. ## Unexpected Solutions Robots sometimes surprise researchers with their behavior. For instance, one robot attempted to sweep items with a banana placed in front of it. Such actions shed light on the development of physical intelligence, which is still largely lacking in machines. The way children explore the world offers important clues for AI researchers. ## Deep Learning and Robotics Generalist AI's founders have backgrounds at companies like Google DeepMind and Boston Dynamics. Traditional robot training required feeding thousands of examples into models, but this method often failed with simple changes like lighting. Generalist is developing a general robot model trained by humans, setting it apart from competitors. ## Specialized Training Data The company uses special gloves equipped with cameras to collect high-quality training data. Humans perform various tasks while wearing these gloves, generating the data needed for training. Unlike others, Generalist builds its AI models from scratch instead of relying on open-source language models, giving it a unique edge. ## Commercial Potential Danfei Xu from Georgia Tech says Generalist stands out in general robot models, praising its high-quality data and scientific rigor. Karen Liu from Stanford University believes Generalist's data approach could be successful. The robots show promise for real-world commercial applications, potentially revolutionizing industries. ## Not Perfect Yet Generalist's models still have room for improvement. On average, a robot completes a demonstrated task only 59% of the time, far from the ideal 99% success rate. It's also unclear how well these skills will generalize across all possible tasks and environments, leaving some uncertainty about widespread applicability. ## A Promising Future Despite current limitations, the potential for robots to quickly learn skills in fields like manufacturing is enormous. One Generalist engineer was thrilled when a robot spontaneously started stacking cups during a late-night experiment. Moments like these highlight the exciting future of AI and robotics, with Generalist AI aiming to revolutionize industries by creating robots that understand the physical world. ## Sources covering this story - https://www.wired.com/story/generalist-ai-robots-learn-like-clever-toddlers/ — --- _Summary written by Brifing from the original reporting. Copyright in the original article belongs to its publisher; read the full piece at the link above._