GEN-1.5 learns robot tasks from one demo

Generalist AI says its GEN-1.5 robot foundation model reached 59% average success across 10 tasks after seeing a single 3–12-second demonstration, with no gradient updates or fine-tuning. In the company’s technical report, the same system reached 83% after 10 gradient steps on five minutes of task data, making the practical story less “robots learn anything” than “a short demonstration can now seed a useful first attempt.” Robot builders should file this as a data-collection signal: the physical-AI funding boom is concentrating capital in deployment-ready systems, but one-shot skills are still modest and brittle enough to demand a human-supervised pilot.