ModelsPhysical Intelligence’s π0.7 asks whether better instructions can replace another round of robot training
A steerable VLA points toward a different model-building advantage: reusing learned skills through richer context, rather than collecting every task again.