taylor-ennen / gpt-streamlit-mvp

GPT-Streamlit-MVP\lesson_plan_streamlit_prototype.py

Here I've provided a simple MVP on Generative Lesson Planning with LLMs such as ChatGPT. ChatGPT-4 Code Analysis through the WebUI was used in producing everything except this readme.

My Observations:

Fine-tuned models could hold higher potential with less dynamic generation being the trade-off in that scenario. This would only be relevant in models of equal size in parameters. Given a general dataset of all possible subjects, a model can be made or trained, to generate lesson plans utilizing pure semantic distillation of the model itself with properly formatted prompting that includes styling expectations.

This is the basic building block of an LLM's capability. While not a novel piece of information itself, the aim of this example is to show how procedural generation can be applied to create learning applications and thus any abstract of such with a similar scaffold or blueprint.

Resources:

Here is the Chat Conversation used to assist in the prototype. Illustrated in the chat conversation is the thought process sideloaded to ChatGPT with minimal time spent in the IDE itself. The decision to publish these conversations and example code was a postscript to the experiment itself.

ChatGPT Share Link containing said conversation: https://chat.openai.com/c/6a55fe17-591d-4819-bab0-74789377815b

You can find the code in its "final state" here: Direct Link

My Biased Opinions:

While not better than a lesson plan curated by a teaching professional this can be a great tool and framework to build simple and quick lesson planning foundations. I'm hopeful that teaching professionals can not only see the benefit but also iterate further in collaboration as the trend of technologies in application ever grows larger and more distant. LLMs will be the next calculator in terms of tooling available to all of modern man.