I am currently trying to build artificial general intelligence.
My other interests include the theory of computation, astrophysics, the Japanese language, and hard sci-fi. You can reach me at email@example.com, or on twitter.
From the hydrogen-red wombs of spiral galaxies like our own to the inner orbits of dying, distant Betelgeuse, the chains of physical law hold fast. The parabolic contours of the smooth surface that dominates the central portion of this ancient automaton still, over eons of dormancy and kiloparsecs of travel, assert their singular, universal, electromagnetic purpose…
We will not have a chance to escape when our societal organizations are deterministic (e.g. government on the blockchain), or worse, superintelligent.
You are taking advice from someone who spent more than three thousand hours studying a complex language spoken on a distant volcanic archipelago in order to watch cartoons more effectively.
I believe that artificial neural networks are profoundly lacking as a substrate for artificial general intelligence (AGI), in ways that scale and architecture cannot fix.
By employing a few dumb tricks, a computer with super-Turing capabilities could prove almost any theorem over the natural numbers, including many that have stumped the best of us for centuries.
Looking at someone’s bookshelf is probably the fastest way to understand them, so here’s mine.