A Columbia student has created a GenAI tool, Curiously, to help educators design personalized AI teaching assistants for their classes.
Yipu Zheng, a Ph.D. student at Columbia University, developed Curiously, a GenAI-powered tool designed to create AI teaching assistants tailored to specific courses. Introduced in a graduate cloud analytics class. The chatbot was trained on course materials and provides support to students 24/7. Zheng’s goal was to address the growing demand for interactive learning tools that are directly linked to course content. Enhancing student engagement and understanding.
One of the challenges tackled by Curiously is minimizing inaccurate AI responses. A common issue with large language models like ChatGPT. To overcome this, the tool uses retrieval-augmented generation (RAG). This combines LLM text generation with an external database of course materials provided by the instructor. This ensures that the chatbot answers questions based on relevant, instructor-approved content, significantly reducing errors or “hallucinations” common in genAI systems.
The platform offers two modes: Concept Assistant and Assignment Assistant. The Concept Assistant helps students understand key terms and course concepts. Whereas the Assignment Assistant supports problem-solving without giving direct answers. Instead, it guides users through steps to solve the problem, based on the instructor’s input, including hints and feedback. This method not only reinforces learning but also discourages dependency on the AI for quick answers.
To further benefit educators, Curiously includes an analytics dashboard that tracks student progress and provides insights on how students engage with the chatbot. As more instructors use the beta version, the tool is continuously refined through data analysis and feedback, making it a promising solution for improving classroom learning through AI.