🔧 Chain of Thought Prompting for LLMs

The advent of ChatGPT and Large Language Models has already affected education. With mixed results and a spectrum of ethical acceptability, students can use chat-tuned LLMs to plan, as a starting point for research, to edit and suggest stylistic or grammatical improvements, or even as a ghostwriter to write assignments.

The well-known non-profit Khan Academy offers its own personalized tutor, Khanmigo, developed in partnership with OpenAI to guide learners using an inductive approach. But despite impressive capabilities in many domains, even the largest and most advanced LLMs exhibit surprising failures, especially in math. If LLMs are prone to glaring mistakes befitting learning students themselves, how can they be expected to act as a trustworthy teaching tool by Khan Academy?