Generalist AI has released GEN-1.5, an AI model that allows robots to learn new tasks from a single demonstration. The model uses a 3- to 12-second demo as a 'physical prompt' loaded into its context window, functioning as the model's short-term memory. After this, the robot performs the task without any additional training. Across ten tests, including opening a jar and pulling money from a wallet, the company reports an average success rate of 59 percent. With ten training steps on five minutes of data, that number increased to 83 percent.
The model can chain two prompts into longer sequences, use demos from simulation, and partly imitate human hand movements. According to Generalist, these abilities emerged naturally over more than eight months of pretraining on interaction data. They were not explicitly trained. Other research teams have demonstrated similar in-context learning, but only for a limited number of task types. Generalist claims to be the first to achieve this across a wide range of tasks.
The tasks demonstrated are simple and short, and all results come from the company itself. None have been independently verified. Source: thedecoder
