The thing I find most suspicious/fishy/smelly about all the current hype around Stable Diffusion and ChatGPT is that it follows, as night follows day, about six months after the bottom dropped out of the cryptocurrency scam bubble.
This is not a coincidence.
Hucksters are chasing the sweet VC/private equity money that has been flushed out of crypto, and AI is the new hotness all of a sudden.
If you're thinking about investing now? Don't: it's too late and you'll be the target of a grift.
Update from comments: I'm SURE it's just a coincidence that training neural networks and mining cryptocurrencies are both applications that benefit from very large arrays of GPUs.
If *I* was a VC I'd be hiring complexity theory nerds to figure out what areas of research are promising once you have Yottaflops of numerical processing power available, then place bets on AMD, Nvidia, and (maybe) Intel going there and start seeding the field and hiring PR firms.
@cstross
I know it makes for a not too satisfying story, but linear algebra is as linear algebra does.
General purpose CPU makers gave up, but the craving for parallel computation is still here...
@cstross that's an interesting theory and I expect some do that, but I don't assume that's the real driver for either. I can't back that up, but I'm skeptical of simple, what many would like to believe takes.
What I know is that outside the VC area more promising, ethically driven technology devt has been going on under the radar since before either and am hopeful one of these will start to come together soon. No GPU's needed, designed for simple legacy h/w and harness unused resources.
@cstross NN hardware reqs are way harder than crypto where they only need to communicate when someone finds a winning hash - training huge NNs needs a *ton *of communication bandwidth both amongst the GPUs and between the GPUs and storage, network/interconnect is as important as compute
The GPUs get fast enough they can't be fed training data or model updates fast enough to actually keep them busy and you can't really effectively use more/better GPUs til you upgrade the interconnects
@cstross I sometimes do reviews of research proposals.
I have not much liked having to determine if the proposed budgets for computing are reasonable or not because of cryptocurrency scammers running up the GPU prices or buying up entire production lines.
I will not like it if "AI" scammers do the same thing.
@cstross VCs have been investing in AI startups for years now - I know because I’ve worked at them for the past 5 years.
What you’re seeing is hype followers, grifters, and the VCs that fall into those first two categories glomming on to the next thing, which is just a questionable, exceptionally resource-hogging subset of AI, not AI as a whole.
@cstross I have a modestly different take on this: It's not that cryptocurrencies were the first such grift, it's that *every* wave of innovation in the last twenty years have a ... Festival.. of people hyping and puffing it.
The venture capital industry took a page from a dark version of Alan Kay's "to predict the future you must first invent it".
They have created a world where the first derivative of capital flows is truth, and that's how they invent.
It's interesting to set the wayback machine to 2005 where you see the VCs of today discovered the wonders of PR:
http://www.paulgraham.com/submarine.html
With social media, the PR has detached from just big firms and turned into a sort of self-pleasuring ball of influence peddling that grew under years of free money into something that wanders around looking for something to latch onto.
The wonder of crypto is that it was basically a complete fraud and hype took it so far.
@cstross Imagine the lesson learned when you see the billions cryptocurrency companies were able to generate in capital investment by every single participant in this scheme.
Where we differ is I don't think every new trend is as big a scam as crypto -- just that the hype machine is so ferocious that they it *Must* be to justify the effort.
@cstross My own take on generative AI is more modestly positive. It has risks, and it's bad at some things (such as being a general knowledge source), but it has a *lot* of applications, and is getting better.
It is *not*, of course, a route to AGI, but the things it can do are sufficiently amazing that when I see the critique that it's just hype, I feel like I am seeing people using it for all sorts of things already.
@cstross Like people are *already using it for porn*, a classic piece of evidence something is happening with a new technology. :P
I can't say anybody ever got boobs on the blockchain.
Charlie Stross
Nemo
onnob
Mark Hughes
Sean Eric Fagan
inhuman resource
Bornach
dr2chase
Richard Welty
Kaiser Bill’s Batman
Michael Busch
An Okay Dad
erisianrite
Brandon
Dave
Borogove
Johann-Tobias Schäg