[overclock351]

Again it's probably my limited scope on the topic that brings me to such conclusions.

I gave a check to the ARC project and it seems like a great challenge (from the standpoint of proactiveness).

What i might not understand is how could a challenge defined like so "AGI is a system that can efficiently acquire new skills outside of its training data." be possible. My mind goes to deep learning, to RNNs, to something that does online training but then again i think that after a while, learning on it's own, wouldn't it introduce noise into it's system? in my mind (again, ignorance speaking here) online training requires some sort of back propagation, some sort of learning on it's own from it's own. The acquisition of new skills must come from either itself or an agent's surrounding environment... i just can't see how this whole system could be stable