Calling something "snake oil" implies that the product only, at best, has a placebo effect (amplified, no doubt, by the sales presentation). LLMs seem to be much more useful than a placebo.
There's also still the risk of the creation of a new economic underclass, if both a) hardware remains too expensive for local inference and b) subscription or pay-per-token based inference also remains expensive or increases in price, then individuals will largely be locked out of the benefits that having access to AI could bring, leaving it purely in the hands of larger companies. People will only get to use and experience these tools through their employer, for the benefit of their employer.
Like in the era of mainframes, before the personal computer was a thing. I wonder if universities will play the same role, though.
The idea that you _must_ have 128 GB RAM and 1 TB ssd in your computer is juts absurd.
Remember... we reached the moon with the compute power of a pocket calculator, and there is no eternal law that says that everything has to be written in javascript.
I don't think anyone thinks that? Unless you are trying to do local inference with bigger models, which isn't everyone's cup of tea but I do think is an important capability.
Outside of that use case though, prices are still ridiculous. 32GB of DDR5 will run you close to $500 right now vs. $80 before the AI build out, and 32GB is what I'd consider comfortable so long as everything keeps being electron and web wrappers (assuming Windows & macOS here for the general population, obviously you can get by with less on a Linux desktop for the most part).
Personally, I'd love to see more software do more with less, and go back to performant native apps but in reality I just don't see that happening, except maybe over here in Apple/macOS land which has always had a decent culture of native apps. For now, the incentives aren't there (for commercial software).
I hope you are right on prices, but I don't share the optimism. We've seen time and time again that once prices go up, they don't always go back down even when the supply crunch is eliminated. Manufacturers realize people adapt to whatever "new normal" prices are at and refuse to lower them. Same thing happened post-COVID supply shocks. Companies have zero incentive to start a price war, even after supply shocks are over.
I think we are stuck with ~$500 RAM for a long time.
Depends how long the RAM correction takes. It is interesting how RAM prices have stifled the creation of cheap laptops capable of running big models. But at the same time, this seems like a second order effect and not the intention.
Choose one:
- You spend 30 hours writing a program to manage data for your hobby. You write it on your personal computer.
- You spend one hour generating a program to manage data for your hobby. You have to lease an H200 behind an API to do it.
Which one will you choose?
I know which one I'm choosing.
I know that many others choose A as well.
A wonderful service known as the web has connected people who choose A with others who choose A and of course with a great many who don’t need to make a choice and benefit from the work of others.
I mourn a world in which few will choose A, because for many to choose B seems to lock us all, tragedy of the commons style, into a worse world.
> AI is better at this than you. You just won't admit it. And it's going to get 10,000x better than you in just a short while.
Where is the 10x (not even 10,000x) revenue? No companies other than those selling the AI itself are seeing it.
It really depends on the cost of immortality. At the very least, it would have a psychological impact that some people may feel is undesirable.
> None of you is writing punch card programs.
> None of you are building vaccum tube logic.
Perhaps none of us, but some people certainly do. We are intellectual creatures. Some of us do things out of pure curiosity. Can we create multinational corporations out of it? Almost certainly not. Can we create businesses out of it? People do so all of the time. There is a market for produce from small farms, hand crafts, heck, even vintage computing.
> None of the things we build today are going to last. Your programs will be meaningless in a hundred years. Probably closer to ten years.
Try telling that to people who are trying to retire legacy systems. Sure, most of them have been modernized. Perhaps they have even been modernized to the point where none of the original code exists. Yet the core ideas still exist since it turns out to be incredibly hard to discard things.
The old ways of writing software will continue, even if they are nowhere near as popular. Call that irrational if you want. I call it human.
Software is better than me at crossword puzzles but that doesn't mean I'm going to optimise my time solving the weekly cryptic
It's just not though. Plagiarizing some shit it stole off github does not make it intelligent.
Edit: just because it's amusing, here's something I'm literally running into right now with Opus 4.8 on "Max" settings being dumb. I asked it to add some C++ code to an existing C++ project for Unreal Engine. It did half the work, then balked, because "it doesn't have a way to compile C++". I just had to tell it "yes, actually, you just need to run the fricking extremely-standard-already-generated-build-script." If I had a novel build system it'd be one thing, but it already knows I'm working on an Unreal Engine project and that the build is completely standard and it still couldn't piece together that it could just run the compiler.
I would not employ a C++ developer that could not figure out how to invoke the compiler.
Exactly, so why do you care how some people build things? It's not that important.
No, you're just really, really shit at programming. You just won't admit it.
(AI, in general, is only impressive when you have no clue about the subject domain.)
Maybe so, but I am better at this when I engage with AI in a controlled manner.
Oh, it was meant to be rhetorical. Because everyone thinks like you?
I don't think you understand what code is. What it does is far less important than how it does it.
Software is bureaucracy and always has been. The discipline is just finally maturing into this role like so many other careers have.
Data centers as infrastructure are very different from DSL rollout though. Much, much more expensive to maintain, with a much much shorter timespan.
If the bubble pops and data centers get shut down because there’s no one to pay the bills, there won’t be much left 5-10 years later in terms of infrastructure.