Microsoft removes guide on how to train LLMs on pirated Harry Potter books

In response to significant criticism, Microsoft has retracted a blog post that appeared to encourage developers to use pirated Harry Potter books for training AI models. The decision came after a lively discussion on Hacker News highlighted the ethical implications of using copyrighted material in this manner. The blog was authored by Pooja Kamath, a senior product manager at Microsoft, who has been with the company for over ten years. It was intended to showcase a new feature that would simplify the integration of generative AI capabilities into applications via Azure SQL DB, LangChain, and large language models (LLMs). To illustrate this feature, the post suggested utilizing the Harry Potter series, describing it as a beloved collection that could engage a diverse audience. According to the blog, developers could create interactive Q&A systems or generate fan fiction based on the Harry Potter universe, which it claimed would appeal to many enthusiasts. However, the post linked to a Kaggle dataset that contained all seven Harry Potter books, which had been incorrectly labeled as public domain. This dataset had circulated online for years and garnered approximately 10,000 downloads, seemingly escaping the notice of J.K. Rowling, who is known for her vigilant protection of the franchise's copyrights. After inquiries from Ars Technica, the dataset was swiftly removed, and the uploader, Shubham Maindola, a data scientist from India, had no known connection to Microsoft. This incident raises important questions about the use of copyrighted works in AI training and the responsibilities of developers and companies in navigating these complex legal waters.