Teachers increasingly set up educational chatbots by describing desired behavior in plain language. A new arXiv preprint argues that how these bots fail in actual classroom use is still poorly understood. To close that gap, the authors introduce EduWild, a dataset of 1,726 conversations.
According to the abstract, the dataset is meant to support analysis of real-world failures and potential remedies. The paper's title promises both diagnosis and repair: "How Teacher-Configured Chatbots Fail in Real-World Use and How to Fix Them."
Because the source is a single preprint with a truncated abstract, this digest is limited to the stated motivation and dataset scale. No comparison with other work is possible from the available material.