This was really interesting. I think a lot of the ideas I’ve had for general architectures have probably already been had by Ben, and I hope we get to talk one day.

I’m especially interested in MeTTa, this ‘maximally general graph-rewriting programming language’. I haven’t been able to figure out why Ben thinks discrete graph queries are so important, but it’s plausible: Computers are good at discrete transforms, the power of them is mostly due to a focus on discrete logic. Machine Learning in recent years has focused on the continuous, the differentiable, but there are things that that may miss. I’d love to see a comparison between goertzelian graphbuilding and whatever GraphCore are doing, if the difference is real.

Also, on MeTTa: “This topos is basically the same as the so-called “Ruliad” previously derived from rewrite rules on hypergraphs, in a discrete physics context [Wolfram’s metaphysics]” Wolfram’s stuff is very neat so that’s pretty cool.

Oh, and in case it needs to be mentioned… We should remember that humanity currently lacks the maturity that is required to build AGI properly, it would be better if we didn’t attempt it until we’ve sorted a lot of other stuff out. Alas, humanity is building AGI, and we have to work with that. We cannot recede from it. We cannot live in denial of it. We must engage with it.