OpenAI and Microsoft documents describe a “doom loop” for the web

Recently unsealed court documents in The New York Times’ case against OpenAI and Microsoft describe internal concerns about how AI systems could damage publishers and the wider web.

Microsoft Director of Applied Science Brent Hecht reportedly called ChatGPT and Copilot’s data harvesting the “largest theft of labor in human history” and said Microsoft’s defense made a “complete mockery” of fair use. Microsoft spokesperson Alex Haurek said the comments reflected one employee’s perspective, were not legal analysis, and did not represent Microsoft’s views. Jordan Usdan described Hecht’s views as academic and forward-looking.

A Microsoft document described an AI content strategy that had created a “doom loop” in which the company’s models and the web could suffer at the same time. The filing also says Satya Nadella acknowledged that chatbots had reduced the need to visit original sources, while an OpenAI representative was unaware of efforts to remove paywalled material from training data.

• OpenAI documents reportedly said GPT-4 had “memorized a ton of data” and was highly capable of reproducing material verbatim.
• OpenAI’s Jack Clark described systems that could substitute for people whose work shapes society’s culture.
• Microsoft was quoted as saying that large language models could destroy their own content supply chain by replacing the sources used for training.

The 92-page filing includes statements from figures including Sam Altman, Greg Brockman, Nick Turley and other OpenAI employees. The Verge reported that Microsoft and OpenAI have sought to distance themselves from some of the claims.