Teaching

Large language models (LLMs) have the capability to not only comprehend and produce content on demand but also creatively integrate ideas from different contexts and cater to individual requirements. This technology may be able to help teachers save valuable time by handling some of the more mundane tasks necessary in teaching, freeing them up to focus on the personal aspects of the job.

---

Lesson planning

Teachers are already using ChatGPT to help generate lesson plans and ideas for activities, but it can be hard to get appropriate content (e.g. for the UK syllabus) without considerable prompting. Can we train a model on custom data to create a tool capable of generating high quality, relevant lesson plans, given just a few parameters e.g. year group, subject and topic?

Oak national academy provides a wonderful resource of thousands of human-created lessons for UK schools. I've scraped the transcripts of those lessons and I'm using Claude 2 to distill those transcripts into concise lesson plans. I'll use this output, along with the lesson metadata, to finetune a model, and then show it to some teachers to see if it could be useful in their work.

[Update: Oak are also starting work on this: https://labs.thenational.academy]

---