LLMs Unleash 'Vicious Cycle' Threatening Web Economics and Referral Traffic

Microsoft and OpenAI have been warned about a 'vicious cycle' between giant language models (LLMs) and the web, threatening both model performance and the economics of websites.

Internal documents from the companies reveal that LLMs can erode their own content supply, as they memorize protected content and reproduce passages verbatim. This has led to a drop in referral traffic for sites like the New York Times, with OpenAI estimating a 60% reduction in search-derived referrals.

Satya Nadella, Microsoft's CEO, has stated that chatbots have replaced search and eliminated the need to access original sources. However, internal alerts at OpenAI mention an 'existential threat' to publishers like the New York Times, with one employee describing data collection by ChatGPT as 'the greatest theft of labor in human history.'

Microsoft has distanced itself from comments attributed to its executive Brent Hecht, stating that they do not reflect the company's position. OpenAI, on the other hand, is focused on generating commercial revenue from AI, with Greg Brockman referring to the 'gazillions' of dollars potentially generated.