From folding boxes to fixing vacuums, GEN-1 robotics model hits 99% reliability (arstechnica.com)

Generalist has unveiled its latest robotic model, GEN-1, which boasts a remarkable 99% reliability in executing a wide array of physical tasks that once required human dexterity, such as folding boxes and servicing vacuums. This advancement signifies a pivotal leap in the AI/ML community, as it demonstrates the potential for machines to not only replicate but also innovate in tasks traditionally constrained by human capability. Unlike its predecessor, GEN-0, GEN-1 not only utilizes enhanced training data but also integrates sophisticated improvisational skills, enabling it to adapt to unexpected challenges in real-time. To enhance its learning capabilities, Generalist employed "data hands," wearable devices that record micro-movements and visual inputs from human demonstrations over half a million hours. This extensive dataset has enabled GEN-1 to achieve its new performance metrics with remarkable efficiency, adapting its training to specific tasks within roughly an hour. GEN-1’s ability to improvise based on previous experiences marks a significant evolution in robotics, moving away from rigid programming toward a more fluid, adaptable model capable of tackling diverse and dynamic environments.