[two_in_one]

While I agree with this emotional post there is one nuance. Neural networks aren't intelligent, brain is. And that's where we want to be. Checking gradients and studying filters can get us only this far. So, using brain as inspiration looks like a good option. There are other, but nobody knows where next breakthrough will be. Like nobody knew five years back that transformers are so powerful. My guess next step to AGI will be a complex modular multi-modal system. With hierarchy, workers and controllers, complex signals.. Sound familiar? Brain is sort of it. This is need for embodied AI, obviously. But, interesting thing, it's needed even for body-less AGI too. I.e. AGI is not a big calculator (!), it's more like real-time system. One reason is that full search is impossible. So, in many cases requests will be like 'give the best answer you can find in 4 seconds'. 'and keep looking'. So far we have only real-time dumb robots and NN big calculators. And brains, of course.