AI training data is expensive, making it best suited for well-funded tech companies. This is why Harvard University intends to distribute a dataset containing around 1 million public-domain books from many genres, languages, and writers, including Dickens, Dante, and Shakespeare, that are no longer copyright-protected due to their age.

The new dataset is not yet available, and it is unclear when or how it will be provided. However, it incorporates books from Google Books, the company’s long-running book-scanning effort, so Google will be participating in the release of “this treasure trove far and wide.”

Harvard originally teased the Institutional Data Initiative (IDI) in March, describing its plans to provide a “trusted conduit for legal data for AI.” However, little has been heard from it until its formal introduction today, which confirmed that the IDI had financial backing from Microsoft and OpenAI.

According to Greg Leppert, executive director of the IDI, the dataset is intended to “level the playing field” by making such a massive dataset available to everyone — from academic laboratories to AI startups — who wants to train large language models (LLMs).