As Meta focuses all of its energy on its generative AI efforts, it's considering whether people should pay for access to higher-quality, more instantaneous training data to improve its tools, and it's turning its attention to the news industry.
Teams at Meta (formerly Facebook) are discussing internally whether to strike new paid deals with news publishers to offer deeper, more robust access to news, photo and video content, according to two people familiar with the discussions, who asked not to be identified so they could speak freely without fear or retribution.
The team discussing access to news content includes the leaders of the partnerships, product and legal teams. Meta may need such access to make its generative AI tools such as Meta AI more effective for users and to better compete in an increasingly competitive market for generative AI search tools and chatbots, the people said.
“Metha may be forced to pay someone,” one of the people familiar with the matter said.
Meta has not formally approached any news organizations about licensing or accessing its content; internal discussions are still in the early stages, according to a person familiar with the matter. But if the company decides to do so, any deals for access to data for model training would be separate from past deals in which Meta paid publishers to link to their content on its sites. BI reported that Meta has made a major shift away from its previous deals with the news industry in the past 18 months, effectively cutting $2 billion from its news division's budget last year alone.
A Meta spokesman declined to comment.
Meta CEO Mark Zuckerberg claimed earlier this year that the company has access to more proprietary data for training its Llama large-scale language model than Common Crawl, the vast amount of scraped web data that is widely used (including by Meta) to train AI models.
Still, there are concerns within the company about the quality of the data it has, one of the people familiar with the matter said. User posts and comments on Facebook and Instagram aren't necessarily the kind of high-quality training data that generative AI chatbots and search tools need to produce quality output — books, news articles and essays are.
If Meta chooses to rely more on its own data, or is forced to, it could find itself at a disadvantage again, falling behind rivals such as Google and OpenAI. Zuckerberg has previously said he doesn't expect a generative AI boom.
Shortly after generative AI gained widespread public attention nearly two years ago with the release of the ChatGPT chatbot, news organizations and other websites began blocking automated bots deployed by Common Crawl and OpenAI to continuously scrape content for free. The U.S. Copyright Office is considering new rules targeting generative AI. Without free and ongoing access to news publishers' content, responses to user prompts about current events from Meta AI may be more limited, outdated, or inaccurate.
Now, the main rival tech companies in the fierce generative AI race are already signing deals with news publishers and media outlets for more access to the content used as training data for their models: News Corp. has a deal with Google; the Financial Times has a deal with OpenAI; as have the Associated Press, Dotdash Meredith, and Axel Springer, parent company of BI, Politico, and several European publications. OpenAI's largest investor is Microsoft, which also has its own partnership with Axel Springer.
Still, many publications have yet to strike deals with AI companies, including The New York Times, which sued OpenAI after it couldn't agree on the terms of a licensing deal. The Times also had Common Crawl remove data it had harvested from the publication. Most news publishers are open to licensing deals, but only because “anything is better than nothing,” said one person familiar with Meta's deliberations.
Are you a Meta employee or just want to share some tips and insights? Contact Kali Hays at tech@meta.com. email address Or in a secure messaging appsignal Please call us at 949-280-0267. Please call us using a non-work device.
