OpenAI and Microsoft Acknowledge LLMs' Negative Impact on Web and Labor

On Thursday, an unredacted court filing revealed that Microsoft and OpenAI acknowledge the detrimental effects of large language models (LLMs) on the web and labor markets. A Microsoft executive described the situation as 'an astonishing theft of unprecedented proportions,' highlighting that clicks to news sites on Bing have fallen by more than 90 percent due to content being appropriated by LLMs. Internal documents from Microsoft refer to this phenomenon as a 'doom loop,' indicating that their AI content strategy could hurt both model performance and the broader web ecosystem.

Jack Clark, OpenAI's policy director, noted that LLMs are creating systems that substitute for the labor of individuals who contribute to cultural content. This aligns with claims in Microsoft’s policy documents that LLMs significantly disrupt the employment of those who generated the data used for training these models. Furthermore, the documents suggest that 'no matter how prominently we show the links, users won’t click,' indicating a fundamental issue with user engagement stemming from LLM usage.

The implications of these admissions are profound for AI practitioners, as they highlight a critical need for addressing the sustainability of content ecosystems and the ethical considerations surrounding AI-generated outputs. This follows ongoing discussions about the regulation and safety of AI technologies, as seen in previous debates among industry leaders.