Nowadays, it’s quite controversial to say that you like and use LLMs. There are lots of discussions about ethics, privacy, and slop. All valid.

One of the chief complaints against modern frontier models has been the massive scale of data scraping. Github repos, pirated books and magazines, forums and posts.

I haven’t been buying into that too much based on some similar prior controversies and resulting court cases involving IP law and big tech. Which isn’t to say it’s all sunshine and rainbows. But it seems to me that the hype is going more towards an LLM-powered, AI-employee which could be the biggest branding blunder of modern tech.

Disclaimer! I’m not a lawyer.

Transformative use

Say, you’re Google, you show people excerpts from books on Google Books. People can get an idea before they buy or rent the book. You get sued for it. You argue that it’s transformative because showing short, interesting passages is inherently different from buying/reading/selling the books. It’s ruled in your favour.

Say, you’re Anthropic, you pirate a lot of books. You combine it with lots of other data, and train a LLM useful for all kinds of purposes. You get sued for it. You argue that it’s transformative because training an LLM is inherently different from buying/reading/selling the books. It’s ruled in your favour.

Maybe not to everyone’s liking but quite based rulings in my book (no pun intended). You create something entirely new that doesn’t distort the old market — fair use.

AI employee

If you say that modern LLM pricing doesn’t make sense, providers are burning investment money to subsidise token pricing in order to capture the market before they rug-pull everyone who’s already locked-in, then it’s a pretty popular opinion.

Karen and her 20 bucks/month plan doesn’t really make the providers any money, so someone has to foot the bill. Even if every office worker runs simple agents (where writing software is probably the majority of the market right now anyway), it might not balance your sheets. What’s the end game?

Internet people have figured it out — LLM providers will sell you LLM-powered, AI employees. They are going to augment or replace many human workers. E.g., Anthropic wants to sell you a security consultant.

If taking licensed code to train a model is fair use because training an LLM is transformative, then how on Earth are you going to convince people that selling an LLM-powered software engineer and displacing a bunch of workers is still fair use?

Not so fair use

You could do some mental gymnastics about growing the pie instead of sharing it but no one buys it right now. The sentiment is pretty negative, people are jaded. What is more, layoffs in the last two years have been largely, albeit falsely, blamed on the AI-driven efficiencies. LinkedIn is full of lunatics posting about their latest slop workflow. As far as marketing goes, LLMs ain’t it right now, and AI-employees would make it even worse.

Ultimately, it isn’t up to the providers to decide what businesses will do. And they typically maximise shareholder value. But if the marketing was about cutting the average person’s work hours by 30% without any trade-offs, no one would mind.

They’re flying too close to the sun.