Joe Rogan Experience #1342 - John Carmack

He talks about AGI at [this video] I wonder if all the really smart people who think AGI is around the corner know something I don't. Well, clearly they know a lot of things that I don't, but I wonder if there's some decisive piece of information I'm missing. I'm a "strict materialist" too, but that doesn't mean I think we can build a brain or a sun or a planet or etc within X years, it just means that I think it's technically possible to build those things. I don't see how we get from "neural net that's really good at identifying objects" to "general intelligence". The emphasis on computational power also makes no sense to me. If we had infinite compute today , what steps would you take to build AGI? Does anyone have any good ideas about that? Sometimes I wonder if AGI (and the concept of a "technological singularity") isn't just "intelligent design for people with north of 140 IQ". Maybe really smart people tend to develop a blindspot for really hard problems (because they've solved so many of them so effectively).

They have a very good thing going. Perhaps there is no great reason to bite off so much at one time. They can take their time and do that later if it makes enough sense. I would expect it would require a very substantial effort to rebuild their platform in a different language. If you're 75/100 of where you want to be on performance, it can be easy to lose immense amounts of time chasing a 95/100 type ideal performance outcome when you can maybe far more easily get to 90/100 by making eg straight-forward caching improvements to what you already have and not have to rewrite all of your code. Good enough is almost always underrated in tech. People destroy opportunity, time, money, and entire businesses chasing what supposedly lies beyond good enough. John Carmack has a good example of this in his Joe Rogan interview [1], in how id Software burned six years on Rage, making incorrect (in hindsight) choices that involved trying to do too much. He regrets his old standard line and approach that it'll be done when it's done. He wishes they had made compromises instead and shipped Rage several years earlier. That's a pretty classic storyline in all of tech, taking on far too much when 85% good enough would have worked just as well most likely. [1] [this video]