Supervisor
Dr. Mehmet Kerem Turkcan
Description
With the rise of LLMs following ChatGPT's nascence, the exploration of potential chatbot intelligent tutors solutions have been further promoted. However, a core problem with LLMs is their ability to produce inaccurate results, often "hallucinating" beyond what is fact and what is fiction. We aim to mitigate this problem by studying the effect of including instructors' information in GPT-4 prompts. This forces GPT-4 to revisit its past encoded knowledge and suggests it to return information that more closely aligns with the given supervised information.
Paper Link
Education Data Mining 2024 Conference: Castleman and Turkcan (2024)