Generalist AI
A robotics startup building models that let a robot pick up a new task from a short demonstration instead of hours of training.
Generalist AI works on the problem that has held robotics back for years: teaching a robot anything new normally means collecting hours of example data and retraining, which takes days per task. The company’s approach is to treat a short video demonstration as part of the prompt, so the robot watches once and tries.
Its GEN-1.5 model, shown in August 2026, reported 59 percent success across ten household-style tasks from a single demonstration, rising to 83 percent with five minutes of extra data. As with most robotics announcements, every number comes from the company’s own testing and none has been independently verified.