How to Learn Everything Using AI? I write about US stocks, which involves reading financial reports and monitoring macroeconomics daily. When I encounter a completely unfamiliar industry, I have to learn from scratch. Previously, these tasks were incredibly time-consuming; researching and asking others would take up most of the day. After using AI Agent for over a year, the first change has been in my learning style. 1. Interest and Motivation are the Foundation Whether this applies universally, I'm not sure, but for me, spontaneous curiosity and self-motivation allow me to persevere longer. Or rather, I can't use the word "perseverance" because I naturally enjoy the process; it doesn't require any willpower, I simply want to learn. If I'm forced to learn by external forces, such as for exams or work, I quickly give up once the deadline is over, or even during the process, I can only barely achieve the average, adopting a "just getting by" attitude. Of course, interest and self-motivation don't guarantee success, nor will they magically appear from the start. Things that are difficult to learn are inherently slow to learn, but taking it slow can actually be faster. 2. Establish a Positive Feedback Environment Positive feedback keeps you interested and motivated, and small, continuous rewards provide emotional value. This is where AI comes in, because its feedback is incredibly fast. For example, if I want to explore a topic, perhaps related to tech or business—fields I haven't studied in school—AI has. I can repeatedly say, "This is too difficult, I don't understand. I need examples, diagrams, and easy-to-understand explanations, starting with the simplest concepts." AI is perfectly suited for this. In real life, it's difficult to get someone to explain the same thing repeatedly, but with AI, we don't have this concern. Without AI, if we encounter something we don't understand, we can only wait for an expert to explain it or search for information ourselves. This extremely rapid positive feedback is something only AI can provide. And when I ask it to generate a question and explain it in my own words, I clearly know what I didn't understand and what I understood well. At this point, AI will also fully encourage me, giving me a lot of positive feedback. This establishes the learning cycle. 3. So how exactly do we utilize AI? Below, I'll break down the specific methods, from simple to advanced. Simple Version: One Chat Window is Enough Free chatbots are already powerful enough; with just a slight modification to the prompt, you can get excellent results. Let's say I'm completely unfamiliar with a certain field or concept, such as "What is an Agent? What are Agents used for? Why are others using them? What can I do with them?" These are the questions that naturally pop into my head when I first encounter a new concept. First, I generally break it down using the "why - what - how" structure. Second, ask only one question at a time, don't bombard them with several. Using this example, you can start with "Why are Agents so popular?" The answer to this question is simple and easy to understand; you'll see some interesting phenomena, and then you can introduce "what," that is, what it actually is. Third, tell the AI your personal background. Free chatbots have no idea how much you know about the field. When you ask them to explain, their answers won't be personalized or consider your background. Therefore, you need to specify in the prompt: "I know nothing about this, please explain in simple terms," etc. Fourth, you can also have the AI use the Feynman Learning Technique to generate questions for you: based on what you just discussed, give it a few questions and ask you to explain it again in your own words. Advanced version: First, get a map, then feed the AI the information. After passing the initial hurdles, if I want to learn more deeply and systematically, before starting, I'll have the AI draw a "knowledge tree" for me in this field: what is the main trunk, what are the branches, and where does what I'm learning now belong? If the field is so unfamiliar that even building a tree is impossible, I'll have it start with its history. Something is always simple when it first comes into being; by following its gradual development into complexity, its current form becomes easier to understand. With the map, I'll have the AI find high-quality books or videos to explore. During the learning process, I also feed it these materials to provide ample context, so I can get answers to my questions whenever I ask. In-depth version: Building your own learning system I intensively use the Agent tool and Obsidian (a personal note-taking software) together. I have a markdown file containing my personal background and a Tutor SOP file, outlining my preferred teaching methods. Each time I ask it to explain certain content, the AI will read these files, supplement the context, and teach me in my preferred way. The rules in this SOP were added after repeated attempts to "get stuck." A few examples: The first is: keep me motivated, keep me going, increase the curiosity. A maximum of two concepts per lesson. The third is saved for the next lesson, ending while the curiosity is still building. Other basic rules include using more examples and diagrams. When I'm interested and want to delve deeper, I also have it read the Feynman Learning Technique's Skills to generate questions for me. I also use Matt Pocock's "Teach Skill" to create web-based lessons using HTML code. I customize them with more interactivity, exercises, and reading materials. It breaks down a topic into several lessons, forming a small set, like 4-5 lessons. Completing a whole set is very rewarding. I once encountered a paper on AI that I could barely understand. Amazingly, the AI kept encouraging me, saying I could do it. I happened to see a section about breaking down out-of-distribution knowledge into in-distribution knowledge. I thought, "This is it!" I'm having the AI break down this completely outside my knowledge distribution step by step, making it understandable. After about a dozen rounds of this, to be honest, I've almost forgotten what the paper actually said, but I still remember the methodology of breaking down knowledge I don't understand and then internalizing it step by step. In summary: the core is actually very simple. You need to like it, be interested, utilize AI tools, ask questions when you don't understand, and continuously receive positive feedback. Naturally, we can learn anything! Here's a prompt you can directly copy: I am [your background, e.g., have absolutely no programming experience], and I want to understand [concepts]. Please explain in the following order, one step at a time, and stop after each step to wait for my response: 1. Why: Why is it so popular recently, and what problem does it solve? 2. What is it: Explain using a real-life example, without jargon or formulas. 3. How to use it: What can someone like me do with it? If I say "I don't understand," please use a different example, not just a different way of saying it. After explaining the three steps, I will be given two questions to answer in my own words, and I will be asked to point out any misunderstandings I made.