Another week, another uneasy deal between Big Tech and the publications covering its players. OpenAI announced Monday that it had struck a deal with the Financial Times to license the newspaper’s journalistic output as training data for its large language models, which in turn power chatbots like ChatGPT.
The FT is the latest outlet to agree to terms with OpenAI, which has already struck deals with publishers in the U.K., U.S., France, Spain, and Germany, among other countries. In what’s been an especially brutal stretch for the media sector, these licensing deals offer newsrooms at least a sliver of hope.
“The deal is good news for the FT and perhaps an inspiration for other large news media interested in striking a deal for incremental additional revenue,” says Rasmus Kleis Nielsen, director of the Reuters Institute for the Study of Journalism at the University of Oxford.
While particulars of the deal haven’t been released, it’s good news for media outlets that OpenAI is willing to cough up the cash, and OpenAI’s latest deal comes as tech firms are grappling with a dearth of good training data on the market now that LLM makers have already hoovered up most of the internet—and have been locked out by “do not crawl” demands by the remainder.
Past research suggests that AI companies could run out of training data by 2026, while a study led by the late University of Cambridge researcher Ross Anderson warned that in the absence of high-quality human-generated training data, LLMs could experience “model collapse,” wherein they’re trained on their own outputs and start to malfunction, much in the way inbreeding affects mammals.
The FT news also comes at a time when AI companies are dealing with a litany of legal issues around their training data. Just this week, eight newspapers owned by the hedge fund Alden Global Capital sued OpenAI and Microsoft for allegedly using their content in training data. “I see this as an approach of tech companies to alleviate some of the copyright infringement they risk,” says independent researcher and consultant Lukasz Olejnik.
However, Olejnik is interested to see whether any comments he provides to FT reporters (he’s been a source there for more than a decade) could change ChatGPT’s perception of him, and therefore what it generates when asked about him. Indeed, a data protection complaint filed this week in Austria by Noyb, a data rights campaign group, raises concerns that ChatGPT queries could yield incorrect data about individuals.