Informal Learning Behaviors Observed in Large-Scale Human-LLM Conversations
A large-scale study analyzing 128,569 naturalistic human-LLM conversations found that informal learning behaviors, such as cognitive engagement, occurred in 31.9% of user turns, while deeper constructive engagement was present in 4.9%. The research identified that scaffolded assistant support is associated with richer, learning-oriented participation, and that these behaviors are selectively and conditionally organized. The findings suggest that human-LLM interactions can foster opportunities for users to reason and construct understanding, rather than merely serving as cognitive offloading.
Why it matters: This research highlights the potential for AI systems to support user learning and cognitive engagement, prompting a shift in evaluation metrics beyond simple answer delivery.
Full story at: arXiv Computers and Society ↗