I love McGilchrist -- I've seen / listened to a number of interviews w/ him and really respect the work he's done. I have yet to read his masterpieces, but they're on my list. (It's such a commitment for me, as mentioned earlier, but I really do want to get to them.)

As part of my obsession with this idea of learning stuff, I've dabbled in some of the accounts that you've described, of savant-like learning. It's been notable to me that of the examples I've come across, the learning in most cases comes with crippling downsides. Borges wrote a famous short story about one such example. A related idea, which you referred to, is the massive process of neural pruning that takes place as babies grow. Forgetting is a crucial part of being in the world. Without a good forgetting algorithm, we are crippled.

Point of all that is that simply "recording" things is a poor analogue for what we usually mean by learning. What we really want, usually, is something more approximated by "synthesis". I think this is consistent with the truth faster examples that @ekzyis and @k00b are giving -- they want, with maximum speed, to integrate the truth with their models of reality. (I think this is what they're asking for, having read the posts.)

I've been noodling on what this would mean in an online-community context, and what it would take to build for it. It would be fun to discuss in some venue.