On the AI Future and the Crypto Future

In my last essay, I argued at length that crypto has very little to offer the world, that the industry has mostly hollowed out and that the main goal of the industry is not to change anything about the world, but to get rich.

Since writing that essay I have had several conversations with friends and colleagues about it. Feedback has been positive! Which is exciting, because writing about the failures of your own industry is risky business, so at least if it reads well, the passerby won’t mind. One common critique that I’ve heard a few times, and must therefore talk about, is that I essentially think of crypto as a dead-end promising a false future that will never be, yet I think of AI as an inevitable future that will take over the economy and potentially destroy humanity. Both are industries that make lofty promises, so how can they be differentiated?

Quite easily, in my view.

AI had a GPT-3 moment

Consider the introduction of GPT-3. Before that, OpenAI was a small research outfit, with no particular renown except its Muskian origin story and a good deal of funding. Those in the know viewed LLMs with great interest, but the common consensus wasn’t that they would prove all that useful for all that much work. In early 2020, there was AI Dungeon which was a fun toy, and there was Coreweave sitting on a stack of too many GPUs, mining Ethereum (it should tell you a lot that they were using their stock for ETH mining rather than AI compute, in those days!), and crypto by then had already gone through some booms-and-busts, so crypto was more in the public consciousness than unheard-of LLMs and AI primitives. Neither was big enough for the typical American to care about either one.

Then GPT-3 was released, everybody in the tech industry noticed the scaling hypothesis was holding steady and the trend line was provably meteoric, and OpenAI’s problem went from having no traction to suddenly having an API product that was so valuable and new that they had no idea how to price it. ChatGPT was years later, in Nov 2022; before that, GPT-3 was a research API that some enthusiast software developers were in line to try, and Gwern was writing about it, but it wasn’t going to change the world in that form; only the prescient noticed anything.

And then ChatGPT came out with enough scaffolding for LLM interaction to make sense to the layperson: before, it was an unstructured, impressively coherent text completion engine. After, it became a personality. Now, three and a half years down the line, it is so capable that it already¹ threatens the viability of many career paths. It will only get better.

OpenAI and Anthropic, and to a lesser extent Google, have revealed to those paying attention that productive autonomous AI agents are the clear future, and are in fact being adopted in some disciplines even today, especially software engineering. One can be forgiven for believing otherwise thanks to the AI slop pervading the internet and the reflexive rejection of the hype on YouTube/TikTok/etc. But the fact is that AI text, code, art and raw reasoning have improved so quickly that, at least on a few-shot basis,² outputs can be indistinguishable from human work. One must remember that, before the summer of 2020, scientific consensus was that AIs with the abilities we have today were decades to a century away. The goalposts are now moving comically fast, but AI is here and marching directly at every goalpost anyone has in store for it.

Crypto has no GPT-3 moment

Crypto, on the other hand, has failed to break out of its own industry except in terms of wild speculation, for reasons I described in my last essay. That speculation has gone quite well for some people who achieved generational wealth as Bitcoin surged from its initial lack of value into the six-figures, from time to time producing big winners. Bitcoin was the progenitor for an ecosystem of countless blockchains, assets and technologies, most of which sound impressive on paper. I see almost no users for most crypto projects, though, and fail to believe that crypto has changed the world in any sense that the layperson cares about. Even with valid use cases in high-inflation countries or those with bad currencies, because crypto can’t be spent in the real world and everyone just wants to convert away form it, we’ve yet to see broad crypto adoption with a serious economic impact in any country; most notably, El Salvador’s experiment with it as a national currency was a failure and has been reversed. You cannot deem an asset a currency of exchange if it’s deemed unstable and nobody wants to hold it.

Stablecoins are more promising, but have the opposite problem where they are useless for the layperson (again, outside those countries where it is difficult to hold a global currency of exchange such as USD or EUR in another form). But Americans and Europeans in the Eurozone only care about stablecoins to the extent they offer a safe, fractional window into treasury bonds, and even then it’s only relevant for decision-making if its yields exceed those of the underlying treasuries, which is often the case thanks to the willingness for many companies offering yields to take a loss with their VC money. This is deficit spending by another name, and predisposes people to find crypto valuable when in fact nothing is happening but a simple subsidy in exchange for user growth or protocol adoption.³

That is a very serious contrast to AI, where the hype is directionally correct. The tech industry has been completely transformed by it in a year, there is all manner of public social alarm about what its capabilities may do to the economy, technologists who have studied the matter for decades like Yudkowsky have been sounding the alarm bells about its growth in capabilities for years, both in terms of the economy and, more importantly, existential risk. That a16z has adopted a crowding-out strategy, and therefore insists on funding a stunning proportion of obviously bad ideas that amount to thin wrappers around OpenAI or Anthropic APIs without any moat at all, is irrelevant in the final analysis.

This does not sound like a technology with a comparable impact on society. I don’t think they are in the same conceptual universe.

For someone who experienced the growth of both cultural phenomena from the 2010s to 2022, one can forgive a shared skepticism of both, but it ought to be clear now that the trajectories of each have diverged severely. Both spheres have subjected the general public to hype cycles and enthusiasts who can hardly talk about anything else. Both demographics have the potential to annoy laypeople who don’t care about the topic, and both claim far-reaching social impacts ahead for their favored technology. But one industry is constantly scavenging for relevancy and the other has its longest-tenured voices (again, Yudkowsky & Soares, MIRI generally, etc.) sounding alarms about the subject, trying to slow AI development progress down, foster international coordination to limit the downside risk and existential risk of the technology, and so on, as data center investment for AI props up the US economy and capabilities redouble at terrifying speed.

Meanwhile, the longest-tenured voices on the subject of crypto are now elite business-owners who stand to personally gain from the hype cycles which readily take to pump-and-dump schemes etc., and who are mostly trying to carve out a viable legal apparatus for themselves before the end of the current presidential administration (and opening prediction markets). As far back as 2018, serious financial institutions were spinning up crypto teams to try to find use cases for the technology, and such teams have long since spun down. The most well-capitalized crypto startups outside the exchanges, like Blockdaemon, struggle to find any reliable product-market fit. Even the exchanges themselves are diversifying their product lines, trying to find another big winner now that nobody credibly expects crypto prices to boom enough to create generational wealth for anybody, relative to the tech stocks they’ve correlated with since 2021.

The notion that the crypto and AI industries are in similar situations is Obvious Nonsense. One is the future, whether we like it or not, to the point where I and many other people are actively concerned about the AI alignment problem and its incumbent existential risk for our civilization, and the other is trying to manufacture a raison d’etre through law and regulation.

The Actual Future

All knowledge workers must now have a plan for dealing with the growth of AI in their industry, with the threat of being obsoleted by the technology looming over everyone. That does not sound like an irrelevant technology. By contrast, as I alluded in the conclusion of my last essay, I see no reason to think carefully about crypto anymore, no serious social, cultural or economic questions that can be answered by a detailed analysis of it, except perhaps whether it will end up in the flow of funds when I make an international payment and how much political corruption will remain possible on its payment rails.

I encourage everyone to read If Anyone Builds It, Everyone Dies. The book is probably correct in its core thesis and, even if not, its ideas are essentially required reading from everyone on Earth today on the economic impact of AI alone.