Babies v climate change; AI v IP; bonds v world

That's right. We have looked at interesting numbers in the news this week and we are here to tell you all about them on today's episode.

How to train your AIs on books and then get sued for it.

Okay, indicators of the week starting off. Darian, what you got?

Tell me about it.

and belches and farts just tootin everywhere.

Right. And countries have been trying to limit global warming to 2 degrees. So what you're saying 0.1 degrees warmer is not a huge increase in the scheme of things?

I mean, we seem to be on the path to having fewer babies anyway. Which is actually the topic of an upcoming episode next week.

Speaking of upending, let's talk about something that is going to upend potentially our whole industry, which is the world of AI. My indicator has to do with Anthropic, which is the company behind the AI chat software Claude. And the indicator is 7 million. That's the number of pirated books the company downloaded into its own internal database a few years ago. And since then, the company has used some of those pirated books along with books that it purchased to train its large language models. And this, of course, did not sit well with some of the authors of those books. What these authors did was they sued Anthropic for copyright infringement. And this week, a federal district court in San Francisco ruled in the company's favor. So the court said using a lot of these books to train large language models is actually okay under a legal doctrine known as fair use. Fair use is this long standing exception to the general rule that copyright infringement is a no, no.

Basically, the judge said that what Anthropic did qualifies for the fair use exception, in part because it was transformative, another legal term. And sure, Anthropic didn't get the author's permission, but by using the books to train an LLM, Anthropic created something really different from the original. And that helps get it to that exception.

big battle that is right. And actually shortly after there was a second victory for an AI company, a US District judge also in San Francisco ruled against Authors in favor of Meta, although it's worth noting that the judge in that case said that training models on authors works could be considered unlawful in some circumstances. So it'll be interesting to see how these cases play out.

So in basic terms, what does this mean?

We just talked about this the other day on the show. We talked to a lot of listeners who spend a ton in the past few months just to get ahead of tariffs and then pulled back.

This episode was produced by Angel Carreras and engineered by Kwesi Lee. It was fact checked by Sierra Juarez, edited by Julia Ritchie. Katkin Cannon is our editor and the indicators are production of NPR.