I really enjoy reading Andy's ideas on education--alongside Peter Gray (a psychologist who emphasizes the importance of play for education, https://www.amazon.com/Free-Learn-Unleashing-Instinct-Self-R...) and Piotr Wozniak (invented SuperMemo, https://supermemo.guru/wiki/Main_Page), he has really shaped my perspective on learning. I actually built a last-minute YOLO application to YC on an extremely similar idea--I figure that modern LLMs are capable enough to offload most of the metacognitive aspects of learning onto. Learn drive (a Wozniak term: your natural curiosity) can take you pretty far in a subject, but it's often frustrating to find the right order to learn concepts based on your current understanding and the subject matter. I've previously scoured syllabi on the internet for this, but often what I want to learn isn't really codified in a single course.
I started building a prototype of this idea that I've been very slowly working on in my free time that indexes and uses my notes in emacs for RAG against a locally running LLM. I do think these kinds of learning LLMs have to be run locally, though I've recently gotten a little frustrated because I cannot run a capable open model without my machine's fans turning on.
If you ever plan on writing on what you're working I'd be interested. I have notes in Markdown or Org forms as well as PDF's I can convert to text easily. I'd like to be able to run a small model locally and tune it with the content i have in the future.
I started building a prototype of this idea that I've been very slowly working on in my free time that indexes and uses my notes in emacs for RAG against a locally running LLM. I do think these kinds of learning LLMs have to be run locally, though I've recently gotten a little frustrated because I cannot run a capable open model without my machine's fans turning on.