Comment by godelski
3 months ago
Looks like an experiment to me. Which is fine. Why not play around? A NN based computer is something people have been contemplating for awhile. Though it seems more like a solution looking for a problem to me ¯\\\_(ツ)\_/¯[0]
> what personal computing could be when designed from first principles.
Actually this bugs me more. I really dislike how frequently people claim "first principles". First principles are those that cannot be reduced. It's often used not in this way and all that accomplishes is people tricking themselves (or worse, tries to convince others) that these are the simplest components. We use first principles in subjects like math and physics but honestly 99.9% of what we work with in those domains are not through first principles. Maybe if you start with axioms in set theory or if you're in physics trying to derive a ToE (what those first principles are) then you're not down there.
It bugs me because deriving first principles is an extremely complicated task. Its also a very beneficial exercise, especially when trying to build things from the ground up. Constantly asking yourself how this can be broken down even more. First principles are not where you start. Having them should demonstrate a large amount of work having already been done and deep thought into what you're doing.
When clicking on that section I see nothing that looks like first principles. I see really a manifesto and much of which is actually difficult to distinguish from current computing.
It's a nice manifesto, but also seems too vague and naïve. Though the latter is generally a feature of manifestos, not exactly a fault
[0] to me, AI being built in should look more like AI being used like physics informed networks. Which are more using the NN for optimization. You could definitely abstract out more and this would be cool and interesting to see. But from the sound of it, it seems more like they're just putting LLMs in
I got the impression it's just a new OS with a different kind of alternative to the internet. Their intent is to have their own systems programming language, UI...etc. I think the idea is to start from scratch and not just rebuild on top of the current monoliths. All of that sounds intriguing to me if they can pull it off. The AI part sounds like something they also want integrated with the OS (maybe their own LLM)? I think you're overthinking what they meant by "first principles". I think that is more about...hey...if we start with this RISC computer...can we build a new computing environment from the ground up based on some core principles like it shouldn't harvest your data and serve you ads and that we don't need a web browser...etc.
Maybe I'm wrong though, but I don't think they're interested in inventing their own universe to create an apple pie from scratch. It sounds like you're alluding to that, which I agree would be an entirely different kind of experiment haha. To further that analogy, it sounds like they may be rejecting the modern industrial/factory approach to buying a Sara Lee apple pie and showing...hey...we can just use an oven and some simple tools.