> We also just brute force "autocompletion" with our learned knowledge and combine it to new parts, which we then add to our learned knowledge to deepen the process

you know this because you're a cognitive scientist right? or because this is the consensus in the field?