A billion without a plan

One passage in Karen Hao’s Empire of AI (chapter 5) stays with me more than the chapters around it. In early 2023 a Chinese AI researcher tells her that what OpenAI did could only have happened in Silicon Valley. China has the talent, he says, but no team there would have been given a billion dollars, let alone ten, for an expensive technology without a clear picture of what it would become and what it would be good for. Chinese money followed only after ChatGPT showed there was revenue to be had. Hao takes it one step further: not even Google or DeepMind put open-ended sums into scaling before ChatGPT. It took OpenAI specifically, with its “billionaire origins”, its ideology, and Sam Altman’s fundraising.

Hao uses this against OpenAI’s claim that AGI was inevitable, and she is right on that point. Nothing about it was inevitable. But I read the same fact as a statement about capital. The bet that produced large language models was made before there was a business case, by a small number of people who controlled a lot of money and understood the technology well enough to want it. A committee, a state plan, or a listed company answering to its next quarter would each have asked the question the Chinese researcher’s investors asked, and each would have said no.

That is what concentrated capital is good at. A market economy lets a few people allocate large sums on conviction, and most of those bets fail. The ones that work fund things no consensus process would have chosen. I argue about this often with people who see Silicon Valley’s founders and investors as the problem. I do not think they are villains. They are allocators, some of them unusually good at it, and over the long run economies that reward effective allocation of resources have produced more than economies that plan them. Whether the bet was a good one for everyone else is a separate question, and Hao’s book is mostly about that question.

Two things keep me honest about this:

• The billionaire origin was more promise than cash. Hao reports (chapter 2) that of the billion dollars pledged at OpenAI’s founding, the lab received only around $130 million, less than $45 million of it from Elon Musk. The pledge bought credibility and talent. The money that paid for scale came later, from Microsoft, a company deciding by very different means.
• Concentration cuts both ways. The same structure that funds a bet without a plan also leaves the costs of that bet, the data, the labour, the energy, to be decided by the same few people. That is the part of the argument I have not resolved.

The broader argument, about who carries progress and why what they build spreads, is in who carries progress.

Related: allocation is the same act at every scale, a person, a family, and a bank all betting on the future with limited means.