#llm-training-data

#llm-training-data

Traffic. Focusing on traffic obscures the purpose of AI answers: to satisfy a need on-site, not to generate clicks. AI-generated solutions do not typically include links to branded websites. Google's AI Overviews, for example, sometimes links product names to organic search listings. Thus visibility does not equate to traffic. A merchant's products could appear in an AI answer and receive no clicks.

For the past two years, my colleague Alex Reisner has investigated precisely how tech companies use massive data sets to train their LLMs. He has repeatedly found that so-called architects of AI have relied heavily on enormous databases of copyrighted work to create chatbots and other programs, and has also found that this work is generally taken without the consent or awareness of its creators: musicians, filmmakers, YouTubers, podcasters, illustrators, writers.