Large language models (LLM) may be experts but they have no experience conversing with students. This project addresses that by creating a student model using authentic student data and using it to post-train off-the-shelf LLM. This paper (28 page PDF) describes the development and evaluation of such a tool. "Our evaluations demonstrate that fine-tuning on authentic learning data significantly improves conversational and pedagogical performance – doubling student talk time, improving questioning style, increasing dialogue turns by 50%, and greater personalization of instruction." Via Philippa Hardman.
Today: Total: [Direct link] [Share]
