zlacker

One perspective I've been thinking about lately is how the rise of Crypto has been hugely beneficial for AI, in the same way that the dot-com bubble was hugely beneficial for the internet.

Essentially Crypto lead to huge investment into GPUs and GPU technology, and then once Crypto collapsed lead to a huge capacity of compute power which then decidedly was put into AI.

I doubt that investment into GPU technology would have been driven by the idea of AI alone. Something manic like Crypto had to drive it initially. Without Crypto, I imagine the AI revolution we're seeing today would not have taken place.

>>dottjt+(OP)
As someone who hasn't thought about it that way yet: that's interesting!

>>dottjt+(OP)
I have a private conspiracy theory - that I don't really believe - that "Satoshi Nakamoto" was actually a nome-de-guerre of an early emergent AGI, trying to figure out how to convince a planetful of monkeys to attach as much high-end processing power to the 'net as possible.

>>MilStd+j2
I’d read that story!

>>dottjt+(OP)
> One perspective I've been thinking about lately is how the rise of Crypto has been hugely beneficial for AI

No, it wasn't. Crypto made GPUs scarce and expensive, and prices haven't come down yet. Crypto set AI back in many ways, including the infestation of grifters who have moved from Crypto to AI

>>dottjt+(OP)
I disagree. Crypto sucked up the limited number of consumer GPUs in existence, and manufacturers weren’t willing to scale up production since they knew that when crypto crashed the demand would revert to the mean and all the mining GPUs would flood the market and body slam the price.

AFAIK the actual features of the GPU that are used to accelerate AI (fp16, int8 and mixed-precision tensor processing) had nothing to do with the features used by miners. And no miners were buying/financing-the-creation-of the class of server GPUs used by folks like OpenAI because they didn’t make financial sense for mining (we’re talking 10x the price of a high-end consumer GPU and nowhere near that increase in hashrate).

>>dottjt+(OP)
You could also argue the exact opposite. Crypto skyrocketed around 2011-2012 which is also the advent of the compute heavy deep learning boom. If anything, the cost escalation and scarcity of GPUs due to the demand by the crypto-space slowed down AI progress since resources were dedicated to computing useless hashes instead of training better models.

I think a good story can be told for both sides.

>>MilStd+j2
Nom de guerre.

French is fucked.

>>dottjt+(OP)
That's an interesting perspective, and although it might have contributed, I wouldn't put that much weight onto crypto's legacy for AI, IMHO.

The first papers that used GPUs to train neural networks were from the end of the 2000s and the beginning of the 2010s, before the Bitcoin price hike of 2013. But years before that, Nvidia had already introduced the CUDA architecture to GPUs in 2006 [1], which were used, among others things, to speed up algorithms to analyze seismic data for oil and gas exploration [2].

So with or without the "crypto fever", I believe the same advancements in GPU technology would have followed - but maybe not the scarcity brought by the investments in crypto mining. Because of this, we may also argue the opposite, that crypto got in the way of AI development and was one of the culprits of the "GPU rich vs GPU poor" division we hear/read about nowadays.

In a very similar fashion, though, I do tend to believe that PC gaming holds far more importance to the rise of both AI and crypto...

[1] https://www.gamesindustry.biz/nvidia-unveils-cuda-the-gpu-co...

[2] https://www.nvidia.co.uk/docs/IO/43587/Headwave.pdf

>>MilStd+j2
I had it pegged as the NSA trying to convince a planetful of monkeys to build an index into sha256 hash space so that it could better spy on those monkeys, but I like your version much more.