Well that escalated… slowly.
Over the past two days, both sides of the copyright war between AI companies and the media have received reinforcements. Early Monday morning the Financial Times announced it had signed an agreement with OpenAI for the tech company to use its content to train AI models and serve it up to ChatGPT users in some form.
Just one day later, several newspapers owned by Alden Global Capital sued OpenAI and Microsoft over, well, using their content to train their AI models and serve it up to users in some form — only without an agreement in place. This is exactly the same move The New York Times made in December, adding a major player on the side of copyright holders suing Big AI. (Disclosure: Media Copilot staff have done consulting work for Microsoft.)
Resolving the copyright issue will have the greatest effect on the information economy of an AI-mediated internet. If AI companies can “ingest” anything they can link to on the open web, they effectively hold all the cards. But if they need to pay up for it or risk having their models deemed infringing or even illegal, it could be disastrous for the nascent technology of generative AI.
Given OpenAI’s multiplying legal opponents, you’d be forgiven for thinking the Financial Times partnership is purely a defensive move. It follows similar deals with Axel Springer and the Associated Press. While each of these agreements pays lip service to ChatGPT having access to news articles and serving them to users with links to the sources, so far that hasn’t happened. If you ask the chatbot for the latest news from, say, Business Insider (owned by Axel Springer), it’ll either give you a grab bag of random facts from the site that may or may not be true, and links that may or may not be real. Or it’ll simply refuse to do it.
Make no mistake, though: OpenAI is certainly planning some kind of generative news experience. And, since Google hasn’t spread the search generative experience (SGE) — where it just gives you the answer instead of a list of links — to Google News, there’s still an opportunity to create a definitive product that combines AI and news.
How ChatGPT Will Do News
Here’s how OpenAI thinks of it, according to a person familiar with the negotiations between OpenAI and news publishers: The agreements deal with a news site’s content in two ways. First, it allows OpenAI access to the site’s archive of everything it’s published as training data for new large language models (LLMs). AI companies were doing that anyway, but at least with an agreement in place the publisher is getting paid for it.
Leave a Reply