I think it’s because humans have multiple neural networks that connect together. Humans have a neural network to understand sight, smell, sound, touch, and taste that all are combined to create one neural network that is interconnected within itself. If you took a human brain that had no experience of anything, and could not process any of the 5 senses and was only able to process language than it would take that human brain billions of texts in order for it to be able to write sentences down. If GPT-3 had a neural network for all 5 senses and you taught it information based off of all of those senses than it could make connections that were previously impossible. GPT-3 will take a word and connect that word with other words to see how it fits into the context of a sentence but humans will take a word and see how that word connects with each and every sense and that takes less information to learn. A human can learn language faster by taking the things they see and connecting them with what the language they are learning. If a human could not connect their sight with language than learning that language becomes much much harder. So the challenge we face right now is learning to connect neural networks with other neural networks in the same way a human connects their neural networks.