OpenAI is reportedly putting as little as $1 million and perhaps only up to $5 million on the table in an effort to strike deals with news publishing firms to use their content to train its large language models.

This might seem like a paltry amount, given the rapid rise of the company’s flagship LLM, ChatGPT, but that might depend on the terms of those deals. The news comes from the website The Information, which said today it had spoken with two executives familiar with the matter. The report states that the company is currently negotiating with about a dozen media companies.

In December, OpenAI said it had partnered with the publishing giant Axel Springer, the first partnership the company has struck with such a publisher. Axel Springer is the German multinational mass media company behind such brands as Business Insider and Politico. The financial terms of the deal were never made public, but according to those two executives, it was in the range of tens of millions of dollars.

Meanwhile, Apple Inc. is currently in the race to develop generative AI and has inked deals with publications such as Condé Nast, which owns Vogue and The New Yorker, as well as NBC News and IAC, which owns The Daily Beast and Better Homes and Gardens, worth in the region of $50 million. According to those executives, the reason for Apple’s fairly large payment is that it wants to use the content more broadly than OpenAI wants in its deals.

Where training generative AI is concerned, it appears there’s no more free lunch. The New York Times, CNN, Reuters and The Verge’s parent company, Vox Media, recently blocked OpenAI’s GPT crawler from accessing data. In December, The Times sued OpenAI as well as Microsoft Corp. for what it said was the companies’ illegal use of its copyrighted material to train their AI models.

Reddit Inc. last year also clamped down on companies using its content to train their generative AI, while well-known authors have also launched lawsuits against such companies that perhaps presumed the authors’ hard work was free to use. Training LLMs could be about to get more expensive.

Photo: Alex Knight/Unsplash