AI Observatory reveals diverse and sensitive uses of generative AI

The AI Observatory, a project co-led by Anka Reuel from Stanford University’s STAIR Laboratory, has released independent data regarding how people interact with generative artificial intelligence models such as Claude and Gemini.

By analyzing real conversations, the study found that user behavior differs significantly from the professional and productive uses highlighted in corporate reports. The research revealed that nearly 48% of conversations would be filtered out by corporate criteria. These non-work-related interactions frequently involve sensitive topics, including health, relationships, adult content, and harassment.

The study also identified specific model preferences among users: Grok and Gemini were frequently used for information retrieval, Anthropic was preferred for coding, Gemini for social roles, and ChatGPT for academic tasks. Additionally, the research noted that ChatGPT interactions became longer and more iterative when utilizing the GPT-4o model compared to GPT-3.5.