'Pencurian tenaga kerja terbesar dalam sejarah manusia:' Dokumen pengadilan mengungkapkan eksekutif Microsoft memprediksi 'loop kehancuran' AI bisa mengosongkan internet

In Dec. 2023, The New York Times filed one of the biggest copyright lawsuits in recent memory against OpenAI and Microsoft, alleging that its AI models heisted ludicrous amounts of journalistic work from the Times. Two days ago, a court filing (hosted and analyzed by Ars Technica) was newly unsealed with all sorts of revealing quotes from higher-ups at Microsoft and OpenAI (thanks, Tech Crunch).

Perhaps the most shocking sentiments are from Microsoft's director of applied science, Brent Hecht, who wrote in an internal presentation that Copilot's answer engine would "hurt the performance of our models and the entire web at the same time."

Internal data from Microsoft suggests that the engine caused click-through rates for The New York Times search results to drop at rates between 87% to 93% compared to standard Bing searches. For Ziff Davis domains—the company that owns Eurogamer and IGN—click-through rates dropped between 51% and 94%.

Hecht called this dynamic, where AI scrapes written work for its economic value and then stuffs it into easily accessed AI answers, a "doom loop … It is highly unusual that an end-product threatens the economic foundations of its essential suppliers, but that is the situation we have created for our LLM business with respect to its 'content supply chain.'"

The filing also reveals that when a researcher for OpenAI, Nick Ryder, told OpenAI president Greg Brockman about "a hack to get around nytimes paywall" to help scrape its writing, Brockman replied "ah nice." OpenAI internal documents see Brockman writing that LLMs are "very good at any news task."

In a Jan. 2023 Memo, Hecht called this "the largest theft of labor in human history.” The typical defense for all this, which the filing extensively rebuts, is that it should fall under fair use. Notably, Microsoft CEO Satya Nadella said under oath in a deposition earlier this year that if he'd known OpenAI was using paywalled information to train LLMs, he would have "[required] OpenAI to retrain its models."