Having read the paper, I think it’s very hyperbolic to say “Potentially describing how general artificial intelligence will look like.” This is definitely a potentially interesting way to aid in transfer learning and reducing the effect of Catastrophic Forgetting, but there are other approaches that have already been published in peer-reviewed (one that comes to mind is Learning without Forgetting). This approach needs to be tested on other domains before even making the claim that it’s the state-of-the-art approach to transfer learning, let alone a path to AGI.