2024-04-07

• Been thinking more about LLMs as learning assistants. A big question for me is around “what is done manually and what is done by the LLM?” I think the gist of my answer here is: the core parts of learning will continue to be done by people, the toil should be done by LLMs. Sounds nice enough. But there are some inconvenient truths that need to be confronted. The first is that learning will continue to be effortful. LLMs will not bring us to something like that scene in The Matrix where Neo gets the Kung Fu floppy disk downloaded directly into his brain. No, people will continue to have to do the core of the work themselves. One example is that people will, absolutely, need to continue to read things themselves. I reject visions of LLMs in education that involve them progressively summarizing all the pieces of text or just showing a set of curated flashcards over and over until mastery. No, first things first, people will have to do the reading. If a student can’t or won’t sit down and read for 30 minutes, there’s just not that much you can do for them. The next thing is writing. People will continue to write and/or create things as a result of their reading. People need to actively make sense of the things they read. A common failure mode in my own learning is reading a thing, then reading some prepared summary, telling myself “nice, I remember everything in the summary, I must be good to go,” and then promptly moving on and forgetting everything within a matter of days. No, we have to actually engage with the material in the form of writing about it, drawing diagrams, acting out ideas in the physical or digital world in some form. LLMs can help here by prompting us to create things that we probably wouldn’t think to create or explore. But the active processing of what is read must still be done by learners. And finally, for people who really want to learn a thing, they will have to review the material. Again, this is an area where LLMs can help, by creating first drafts of flashcards/things to review based on the notes or other creative outputs on the relevant subject. But the actual reviewing of the cards/information will be up to the student. LLMs can also probably help a student stay engaged with these reviews better than existing systems, but, again, it is only a helper/facilitator. Ultimately, the core of learning (reading, writing, reviewing) will be done by students. LLMs can remove some of the toil (e.g. translating notes into flashcards for review, creating visualizations of concepts from written notes, making it possible and/or natural to review information using speech, etc.), but the inherent complexity and work associated with actually learning something will remain. Hopefully, in addition to full on removing toilsome elements, the LLM can also make the actual work feel less like work and be closer to fun. But that is a bit of a dangerous game. Many of the learning platforms that start optimizing for “fun,” seem to turn into things that don’t really accomplish much learning (e.g. Duolingo).

---

---

---