At the center of the New York Times’s recent lawsuit against OpenAI is the claim that ChatGPT outputs provide users with “near-verbatim” excerpts from the publisher’s articles.
On Monday, January 8, OpenAI defended its use of NYT journalism to train its large language models (LLMs). It argued the NYT manipulated ChatGPT to generate the supposedly plagiarized outputs cited in its lawsuit.
Hitting back at the Times’ copyright infringement claims in a company blog post, OpenAI accused the newspaper of specifically designing ChatGPT prompts to generate outputs that closely approximate copyright-protected articles.
It said: “It seems they intentionally manipulated prompts, often including lengthy excerpts of articles, in order to get our model to regurgitate. Going even further, the firm speculated the NYT “either instructed the model to regurgitate or cherry-picked their examples from many attempts”.
Lead counsel for The New York Times Ian Crosby, a partner at the law firm Susman Godfrey, told CCN that the complaint was about OpenAI seeking to “free-ride” on the Times’ journalism. He said the company was using it for free to build “substitutive products” without the publisher’s permission. He argued: “That’s not fair use by any measure.”.
In its lawsuit, the Times claims it initially approached OpenAI and its part-owner Microsoft in April. It says it opened negotiations over the use of its articles to train AI models. The publisher claims it sought a resolution potentially involving a commercial agreement and a commitment to “technological guardrails” around AI models.
According to OpenAI, discussions centered around a “high-value partnership” on “real-time display with attribution in ChatGPT.” The firm said talks were still progressing constructively as recently as December 19. It added that it was blindsided by last month’s lawsuit.
While the company failed to reach an agreement with the New York Times, a recent partnership with Axel Springer points to the kind of relationship with major publishers favored by OpenAI.
Covering the use of content published by Axel Springer’s media brands, including Politico and Business Insider, the three-year licensing agreement will see ChatGPT generate summaries of news articles, including otherwise paid-for content.
The complaint claims OpenAI should be held responsible for “billions of dollars in statutory and actual damages”. In contrast, the Axel Springer deal is reportedly worth just tens of millions of euros.