Why aren't any of you geniuses working on AGI? Such as applying some of these NN advancements to agents in diverse 3d virtual environments attempting varied/gradually increasingly complex tasks and learning based on low-level (grounded) sensory inputs?