So Jim is raising the argument we should train models via videos of the task as a superior route to training, makes sense given Nvidia has a robot warehouse sim

So Jim is raising the argument we should train models via videos of the task as a superior route to training, makes sense given Nvidia has a robot warehouse simulator and wants to sell (pickaxe's) to business.