Added OpenAI statement and additional background information.
Update from January 5, 2024:
Tom Rubin, OpenAI’s chief of intellectual property and content, confirmed to Bloomberg News that negotiations with “dozens of publishers” have been “very positive” and are “progressing well.” He expects more deals like the one with Axel Springer in the future.
Training AI models on publishing data is not like the emergence of social media or search engines, Rubin said. AI models would neither replicate nor replace content. The New York Times lawsuit contradicts that assertion.
AI companies negotiating with publishers could be a sign of things to come
The negotiations show that OpenAI – and, inevitably, other AI companies – are willing to deviate from the pure fair use approach they took before the U.S. Copyright Office.
Meta, Google, and OpenAI argued that data training for AI models is a transformative use, and therefore a royalty-free use is legitimate.
The fact that OpenAI, Apple, and Google are now entering into negotiations can be considered a sign that they are softening this stance. Apparently, the companies are not sure that their fair use argument will hold up in court, probably because there is some degree of memorization in these large AI models.
This raises the question whether AI model providers still have a business model when licensing costs are added to the high costs of training and running the models.
In its submission to the US Copyright Office, Meta described the cost of licensing AI training data at the scale required as unsustainable but acknowledged that agreements with individual rights holders are possible. But where is the line?
Original article from January 4, 2024:
Apple to outbid OpenAI in race for AI contracts with media companies
OpenAI is offering some publishers between 1 and 5 million dollars a year to license their news articles to train large language models. Apple is reportedly offering more money, but also wants broader rights to use the content for future AI products. OpenAI is currently in talks with up to a dozen publishers, while Apple is reportedly close to finalizing AI deals with about a dozen media companies offering multi-year contracts worth at least $50 million.
OpenAI recently signed a contract with Axel Springer worth tens of millions of euros per year. According to The Information, Google lags behind OpenAI and Apple in negotiating licensing deals with publishers.
The New York Times Co. recently sued OpenAI and Microsoft for copyright infringement, a move OpenAI described as “surprising and disappointing”. License negotiations appear to have broken down here.