In a public response to a copyright lawsuit filed by The New York Times, OpenAI has dismissed allegations, claiming they are without legal basis and expressed optimism about establishing a partnership with the media giant. OpenAI stated that the Times deliberately omitted some crucial facts and disputed claims that its AI tool, ChatGPT, reproduced New York Times articles verbatim. It argued that the Times manipulated prompts to include excerpts from previous articles. The ChatGPT owner explained that its models do not typically behave in the way the Times insinuates, suggesting that the newspaper either instructed the model to regurgitate content or 'cherry-picked' examples.

The company acknowledged its efforts to reduce content regurgitation from its language models and revealed that the Times had refused to provide reproduction examples before filing the lawsuit. It admitted unintentional content reproduction and removed the Browse feature from ChatGPT.

OpenAI has previously argued to the UK House of Lords that AI systems like ChatGPT cannot be developed without access to copyrighted content, maintaining that AI models require access to vast human knowledge to learn and solve new problems. While it respects copyright ownership and offers opt-outs for including training data, OpenAI argued that training models using internet data fall under fair use rules. The company emphasised the importance of incorporating copyrighted works into AI tools to represent diverse human intelligence and experience and announced that starting from August 2023, website owners could block its web crawlers from accessing their data.

Despite the lawsuit, OpenAI expressed hope for a constructive partnership with The New York Times, citing successful collaborations with Axel Springer and The Associated Press, acknowledging the newspaper's long-standing history and showing respect for it.

Why does it matter?

The significance of this news is underscored by a trend where publishers, artists, and writers increasingly resort to legal action against popular generative AI tools for using their content as training data. Despite such challenges, and the OpenAI's deal with Axel Springer, following a prior agreement with the Associated Press, highlights the growing intersection of AI technology and news publishers. These deals, including OpenAI's potential compromise with The New York Times, also mark a notable milestone in the evolving relationship between journalism companies and AI firms. Unlike conventional practices of supplying data for training models, it introduces a novel dimension by utilising vetted journalism to improve the precision and reliability of ChatGPT's responses.