Generalist AI introduced the GEN-1.5 robotics model, which learns a new task from one short demonstration
Generalist AI introduced GEN-1.5, a model that, according to the company, learns a new task from a single 3 to 12-second demonstration without further training. Without training, it achieves a success rate of 59 %, rising to 83 % with ten training steps on five minutes of data. The results have not been independently verified.
Generalist AI introduced GEN-1.5, a model designed to control robots. According to the company, uploading a short task demonstration lasting 3 to 12 seconds into the context window of the model as a so-called “physical prompt”, which acts as the model’s short-term memory, is enough for the robot to then perform the task without any further training. Across ten tested tasks, such as opening a jar or taking money out of a wallet, the company reports an average success rate of 59 percent. After ten training steps on five minutes of data, the success rate increased to 83 percent.
According to the company, the model can combine two prompts into a longer sequence of steps, work with demonstrations from simulations, and partially imitate human hand movements. The company says these capabilities emerged spontaneously in the model during more than eight months of pretraining on interaction data, without being explicitly trained.
Other research teams have previously demonstrated similar in-context learning, but always for only a limited number of task types. Generalist AI claims it is the first to achieve this across a broad range of tasks. However, the demonstrated tasks are simple and short, and all results come from the company itself – none have been independently verified.
Why it matters
If the company’s claims were confirmed, it could mean that robots can be taught new tasks without costly collection of training data and repeated model training – one short demonstration would be enough. This is particularly relevant for companies developing robotic automation, where retraining for new tasks has so far been costly in both time and money. For now, however, these are company-reported figures from simple and short tasks without independent verification, so real-world usability beyond the presented tests remains unclear.
Relevant practical impact
What this means
For a business
According to Generalist AI, an approach using a “physical prompt” could reduce the need for extensive robot training for new tasks, which could mean faster and cheaper adaptation of robots to new tasks for companies deploying robotic automation.
DevelopmentCheck the original
Event sources
only one source so far · 1 publisher, 1 independent. We count feeds from the same owner only once.