LLMs as Tutors
This is not a grift, I promise lol
Recently I’ve been reading about how LLMs can be used as a learning partner or tutor to explore topics of interest. I decided to explore this possibility for my own purposes; an exercise I thought would also help me further compare tools like Claude 3, GPT4, and Gemini Advanced to see which would be the best for my needs.
Editors Note: I know, I know, who isn’t talking about pRoMptInG and AI these days. There’s so much absolute garbage content out there around AI and LLMs, 99% of which is someone trying to make a quick buck off of a subscription and/or by selling your private data/information. I personally almost feel gross writing about it now, despite the fact that I have nothing to do with any of it. That being said, what follows contains no grifts. I think there are many amazing applications for LLMs that a quiet minority are discussing - and those are the things I want to talk about.
My exploration led me to More Useful Things, the companion site for One Useful Thing, a research blog by Dr. Ethan Mollick and Dr. Lilach Mollick of the Wharton School. One Useful Thing explores the implications of AI for work, education, and life in terms that we can all understand. In my opinion, it is a thoughtful, interesting publication that discusses meaningful impacts AI can have on our lives.
More Useful Things has a prompt library primarily geared towards use in a classroom/learning setting, all of which follow a structured format. Most of the prompts are also tailored for a specific LLM of your choosing.
The instructor-oriented aids can help generate project ideas, quizzes, syllabi, and even structured prompts themselves. I took a particular interest in this last possibility, since it could be used to create my own customized tutor, based on what level of knowledge I have and what learning goals I want to achieve. In this scenario, I am both teacher (in creating the prompt) and student (in using the prompt I created to access a customized tutor.) What I found was that the LLMs asked me questions to generate a structured prompt capable of asking open-ended leading questions. I believe this to be an extremely strong tool for anyone looking to study or learn more about literally anything in a customized manner. [1]
The prompt library’s section oriented for student-use also includes a “General Tutor” prompt, but I believe the extra step described in the last paragraph allows for a more tailored experience. Other student-oriented prompts include a negotiation simulator, a mentor, class reflection aid, and even a scenario where the LLM plays devil’s advocate. I have yet to try all of these out, but I fully intend to revisit them when need arises.
The Dr.’s Mollick have created something incredibly special here. It’s often been said that the internet changed the way we learn forever.[2] The internet gave anyone with a connection access to near infinite knowledge. However, the biggest shortcoming of learning on the current internet is you have to know what to look for and where to find it. Now, using the vast knowledge of LLMs, we can ask questions and be asked questions in an accessible, interactive learning environment from the comfort of wherever we are. On top of this, our tutors can adjust their approach dynamically based on our needs and the ways we learn.
I think we are extremely close to this technology being truly accessible to everyone, and I cannot wait to see how it unfolds.[3] Until then, I recommend checking out One Useful Thing and More Useful Things, and trying the prompts to learn more about whatever your heart desires.
I literally cannot express how important I think this capability is. If we can just reduce the amount of ghosts of misinformation in the machine, so to speak, we can essentially provide everyone with a customized learning aid that adapts to what they need and how they learn. ↩︎
I don’t actually know if I made that up, or just think that. But perhaps I’ll expound on why I think that in a later post. ↩︎
Sources. Outside of the need of general availability, I believe a lack of information sources is the biggest shortcoming of current LLM-based learning. It’s important that we not only be able to follow the rabbit trail of information, but also be able to verify that what we are being told is based in fact. ↩︎
