dailyai.report

The startup Physical Intelligence released π0.7, a foundation model that recombines learned skills through compositional generalization. It mimics how LLMs reassemble text fragments to handle new tasks. While the model shows early promise in versatility, it still exhibits significant flaws. This approach offers a blueprint for scaling robotic dexterity via general-purpose training data.