February 2023 Archives

The thing I find most suspicious/fishy/smelly about the current hype surrounding Stable Diffusion, ChatGPT, and other AI applications is that it is almost exactly six months since the bottom dropped out of the cryptocurrency scam bubble.

This is not a coincidence.

To me it looks very much as if the usual hucksters and grifters are now chasing the sweet VC/private equity money that has been flushed out of the cryptocurrency market. AI is the new hotness, all of a sudden, not because it works but because it delivers panicky investors on a platter.

If you're thinking about investing in AI startups now? My advice is to avoid them like the plague (unless you are absolutely certain that you understand both the technology and the market for the proposed applications): it's too late, and you'll wake up one morning only to discover you've had your pockets picked.

Much of what passes for "journalism" these days is just stenographers feverishly copying the press-releases they're spoon-fed. Real journalism is a niche sector, and unless you subscribe to the exhorbitantly priced newsletters of the high-end analysts who are paid to work full-time studying the sector, what you're seeing on the news websites and in the newspapers is the product of PR firms paid to push AI. And you really need to ask who is paying them.

The AI sector will tick along for a while, generating positive headlines, but there's going to be a crash, sooner rather than later. The Bing/ChatGPT fiasco is just a harbinger of the way deep learning models are going to be discredited in the public eye, as people gradually realize that most of what you get out of a garbage model is more plausibly remixed garbage, and that producing a non-garbage training model requires careful human curation of the sort of giant heap of data that is incredibly labour intensive to prepare. I give it about 1-3 years until the crash. (Although I tend towards optimism: the cryptocurrency bubble took a bit over a decade to implode, but back in 2011 I prediced its demise within, yeah, 1-3 years.)

As for what you should look to invest in?

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. As in, hundreds of thousands to millions of GPUs soaking up entire nations' worth of electricity. (If I recall correctly, the latest ChatGPT model was trained on a supercomputing cluster that turns nearly $2M of electricity a year into waste heat: and it took a couple of months of 100% usage.) And of course, AMD, Nvidia, Intel, and the usual suspects have never imagined paying a PR firm to talk up markets for their latest products.

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 I'd be placing bets on the GPU manufacturers going there. Then I'd start seeding the field and hiring PR firms to push the narrative.

But I'm not a VC, so this is just spitballing.



About this Archive

This page is an archive of entries from February 2023 listed from newest to oldest.

January 2023 is the previous archive.

Find recent content on the main index or look in the archives to find all content.

Search this blog