I’ve been quiet for a while…busy digging deeper into the GenAI world and iterating on tools for schools.
A common use case that I have seen popping up is the ability of LLMs to scan PDFs and provide an interface for interacting with the information contained within. LLMs are excellent at summarizing information, and inferring context, relationships, and insights from data within the document. This mean that extracting key themes, identifying trends, and understanding implicit connections are easily accessible.
A few weeks ago, Momo had an algebra test. As part of preparing for it, we discussed how an LLM could analyze his homework sets and generate possible lists of questions for the exam. Using some simple python code as a wrapper for HuggingFace and OpenAI, I created “Notes Wrangler”:
The app itself is straightforward – it uses Streamlit for rapid development of the interface. The app takes one of more PDF uploads, processes them using LangChain, and then is available for questions. We uploaded homework sets and the results were fantastic. From a PDF of a homework set, we were able to generate various lists of possible questions for the exam.
Something I remember from early in academics was preparing for a test by reviewing class material, especially homework sets, to be ready for possible questions. I recall being particularly proud when I was able to predict a question almost exactly, in advance of the test.
It will be interesting to see how tools similar to this change the way students prepare for tests. Is this increasing productivity by facilitating the process, or part of a slow atrophy of the kinds of skills needed by students? The jury is out, but I suspect not for much longer…