Using Large Language Models to Scale Up Teacher Guides

Many of the most effective interventions to improve literacy and numeracy center on structured pedagogy. These programs aim to illuminate a path toward effective instruction by giving teachers a step-by-step guide for teaching content effectively. The goal is that guides scaffold teaching and build teachers' fluency in an approach grounded in the science of teaching while making using powerful teaching practices feel rewarding and routine. Although evidence on the cost-effectiveness of structured pedagogy at scale is compelling and demand for them is growing, the technical and cost barriers to their development are high. As a result, the adoption of these methods remains limited, and many children and teachers are yet to benefit from this highly promising approach to teaching. A key constraint to scaling up structured pedagogy approaches is that structured pedagogy experts must manually adapt teacher guides between contexts. This has meant that expert teacher trainers have had to spend time on tasks such as adjusting content slightly to account for curriculum or school calendar differences or translating material between languages. 

This Learning for All Initiative research grant will focus on systematizing and automating many tasks using LLMs like ChatGPT. During this pilot project, Inspiring Teachers will focus on determining the feasibility of integrating LLMs into their structured pedagogy development process. They will also work with J-PAL-affiliated professor Jason Kerwin to assess this approach’s impact on program efficiency and quality and prepare for a larger RCT.