Are LLMs a Utility?

Maybe Large Language Models are like electricity. Useful, yes. Revolutionary, sure. But still, just a utility. No one is getting venture returns investing in power plants (people keep forgetting this—see solar power investment returns), and yet all economic activity today relies on the stable, constant delivery of electricity.

Most work doesn't need genius. Literally, by definition. 95.4% of white-collar jobs happen below 120 IQ (two standard deviations). LLMs are already there, more or less. And inference costs? They're in freefall—like Moore's Law, but for AI. So far, they’ve been halving every eight months (or faster) in terms of price per c. Good for users, bad for providers.

So, who wins? Mr. Masayoshi Son would believe that he’s going to be getting outsized returns plowing $40 billion into OpenAI, believing that the winner-takes-all/network effects approach applies here as well. I think a different economic model applies. LLMs are a commodity market. What’s the network effect of ChatGPT? There are no switching costs for me to use Gemini vs. ChatGPT vs. DeepSeek vs. Claude. They are not getting measurably better the more people use them (see DeepSeek). I think economic value is going to flow to applications—specific tools for specific problems.

RegTech is a perfect example. Regulations aren’t a secret; they're just there, in the public domain. The reason why financial institutions spend staggering amounts on regulatory compliance is because the overlay of regulatory compliance on top of business processes is a complex, nuanced task, requiring experience and expertise. Something that can’t be learned or digested from a textbook, but gained through experience. A generic LLM isn’t going to get you there, but specialized applications, that are able to deploy expertise, but at scale? Well, I think that’s where value is going to be.

“Value" has no meaning other than in relation to living beings. The value of a thing is always relative to a particular person, is completely personal and different in quantity for each living human - "market value" is a fiction, merely a rough guess at the average of personal values, all of which must be quantitatively different or trade would be impossible ... This very personal relationship, "value," has two factors for a human being: first, what he can do with a thing, its use to him ... and second, what he must do to get it, its cost to him. There is an old song which asserts "the best things in life are free." Not true! Utterly false! This was the tragic fallacy which brought on the decadence and collapse of the democracies of the twentieth century; those noble experiments failed because the people had been led to believe that they could simply vote for whatever they wanted ... and get it, without toil, without sweat, without tears. Nothing of value is free. Even the breath of life is purchased at birth only through gasping effort and pain.”

―

Robert A. Heinlein

‍