Achieve Mastery with Ease: Employ LLM as Your Personal Socratic Tutor

Paradigm shift of self-study post OpenAI-o1 era

On how to really “learn” things, great scientific discoveries have been made, as perfectly summarized by the book <Make it Stick>. Read it and you’ll likely feel encouraged “Image how well I’ll do in my next study session”. No, that ain’t gonna happen because you need time to get familiar utilizing those procedures, and more importantly, we humans are lazy — we are reluctant to make flash cards or keyword summary for active recall or interleave concepts, as suggested in <Make it Stick>.

No worries an AI Socrates got you back at the rate of less than $0.5/hour (and you’re unlikely to study a full 8 hour a day, converting to $4/day).

Let’s cut to the chase — prompt you could copy paste

Conceptually, you’ll want to include two components in your prompt:

1. A specific topic you want to learn/explore/review
2. A list of principles from <Make it Stick> so LLM applies them to your topic
3. Instruct the LLM to chat with you in “Socratic Questioning” style. This results in natural conversation flow and step-by-step level up of learning.
4. (Optional) Your background and current progress.

I’m still lazy to write #1-#4 myself, so I ask Claude3.5 (or equivalently GPT-4o) to help me write a prompt that I’ll feed to o1 model

Below are some long summary of book on how to study <Make it Stick>. I’d like you to write a prompt. The prompt will do two things:
1) contain core principles of the summary of <make it stick>.
2) ask AI to leverage those best practices and ask me questions like socratic questioning to test my understanding of core concepts of introductory level reinforcement learning.
Summary of make it stick:
… // here I simply copy-paste the summary of the book from the article linked at the start of this article

Let’s appreciate the prompt generated:

Using the core principles from “Make it Stick,” including active retrieval, interleaving, desirable difficulties, elaboration, generation, reflection, and growth mindset, please engage me in a Socratic dialogue to test my understanding of introductory reinforcement learning concepts. Specifically:

1. Ask me questions that require active recall of key reinforcement learning terms and concepts.
2. Interleave questions about different aspects of reinforcement learning (e.g., agents, environments, policies, value functions, and algorithms).
3. Present some questions that create desirable difficulties, such as comparing and contrasting different reinforcement learning approaches.
4. Encourage me to elaborate on my answers and generate examples or analogies.
5. Prompt me to reflect on my learning process and any misconceptions I might have had.
6. Frame the questioning in a way that promotes a growth mindset, emphasizing that understanding reinforcement learning is a skill that can be developed with effort.

Please start with a fundamental question about reinforcement learning and adapt subsequent questions based on my responses, aiming to deepen my understanding and uncover any gaps in my knowledge.

Feel free to copy paste above and replace my topic (reinforcement learning) with your favorite and give it a try. You’ll be amazed I guarantee.

Why o1 Model

Just as Terence Tao evaluates o1 as a “mediocre, but not completely incompetent, graduate student.”, it’s capable to guide you the learner on undergraduate to graduate level study topics, with more structured and in-depth conversation. It’s much better on this than older models.

Tailor-made your own prompt

If you enjoy this approach you should consider save the prompt somewhere and refine it as you use it. This builds more powerful LLM answers and also help you know better of your own learning habits.