←

How I Learn With LLMs

llmlearning

TLDR:

  • I made a flowchart of how I usually learn things with LLMs.

Studying with LLMs has taken some getting used to, but I have settled into a pattern when I use them to learn or build something technical. The flowchart below has the steps I roughly follow. I almost always begin with Wikipedia. The first couple of paragraphs are usually a good initial summary. Then I hit some friction, usually because Wikipedia is not always written in a gentle way. At that point I go to the best LLM at the time and ask it for the best sources on the subject. For something biological or novel I ask for a scientific review. For something more technical or standard I ask for textbook recommendations. I find these types of sources make the best learning paths.

Flowchart of how I start a technical project. Start with Wikipedia, ask an LLM for reviews or textbooks, follow the source, ask an LLM at roadblocks, switch sources if one is not helping, gather notes, then create a final learning document.

I then work through the source, asking an LLM for help when needed, either for worked examples or to fill gaps. If a source does not work out, I move on to another one. I am a visual person and the gif below shows how I think about this loop. You fill in the gaps with back and forths. This works even better with an actual subject matter expert instead of an LLM, but that is rarely possible. Once I am satisfied with my understanding I put all my notes together and do a final write up. That could be a presentation, a digital document, or pen and paper.

Animation of a circle filling in over time. Green areas are understanding that came from sources and working on paper, orange areas came from LLM explanations and examples.

Everyone does this a little differently, but this is what works best for me so far.