Learn Anything with AI and the Feynman Technique

study any concept in four easy steps, by applying AI and a Noble Prize winner approach

Image created by GPT-4o.

When was the last time you stumbled upon a difficult subject to learn? Or when you spent an hour watching YouTube videos on how to better learn things?

There are countless learning techniques to help you digest complex concepts and feel confident about knowing them by heart. And if you’re a student like me who is constantly learning things, you understand the significance of an effective learning approach. One of the simplest one of them is the Feynman Technique.

In this article, I will explain how to apply the Feynman learning method effectively, and how you can use Artificial Intelligence to fill in the gaps of your knowledge.

By the end, you will be able to use ChatGPT to break down complex concepts and master them intuitively and effortlessly in four easy steps!

What is The Feynman Technique?

Richard Feynman was an American theoretical physicist. As part of the Manhattan Project, He played a crucial role in the development of the atomic bomb during World War II. In 1965, he won the Nobel Prize in Physics for his work on quantum electrodynamics. But beyond all that, he was a popular teacher and author of famous books.

Richard Feynman while giving a lecture. (source)

Despite all the impressive achievements, Feynman didn’t believe himself to be intellectually special, but rather an ordinary person who could commit himself to studying hard.

I was an ordinary person who studied hard, there’s no miracle people. There’s no talent or special miracle to study quantum mechanics that comes without practice and reading and learning and studying. — Richard Feynman [1]

Now the Feynman Technique is not directly devised by Feynman but associated with him. Nevertheless, it is inspired by how Feynman believed a subject must be studied.

I couldn’t reduce it to the freshman level. That means we don’t really understand it. — Richard Feynman [2]

Image created by GPT-4o.

Feynman’s Technique

Feynman was famous for his ability to explain complex physical concepts in an intuitive and digestible fashion. He believed that you can only claim you have understood a concept, if you can explain it understandably, to someone who does not have any prior knowledge about it. Nobody could say it better than Feynman himself,

When we speak without jargon, it frees us from hiding behind knowledge we don’t have. Big words and fluffy “business speak” cripples us from getting to the point and passing knowledge to others.

Feynman’s technique for learning a topic can be broken down into these four simple steps:

1. Teach the concept: The most effective method to understand something is by teaching it. Whether you want to imagine teaching the concept to someone else, yourself, or an imaginary child, you must assume the other person knows nothing about the subject. So don’t hide behind and from big words.

2. Identify gaps: Go through what you have been teaching. From the other person’s point of view, try to identify parts of your explanation that are missing, need more work, or are simply not understandable enough.

3. Refine: Using the feedback of the last step, iteratively refine your explanation until you are happy with it.

4. Tell a Story: Now that you have the foundation, fortify it with examples, illustrations, and diagrams. Make your explanation so fluent that you can convey it through a story that is easy to the ears and entertaining to follow.

AI + Feynman Technique = 🔥

You're falling behind if you’re not using AI to enhance your learning process. In this section, I will introduce a very simple method to couple AI with the Feynman Technique by prompting ChatGPT. As an example, I will use “vector database” as a machine learning concept I would like to learn.

Step 1: I will read about the topic and study a basic understanding of it. Then, I developed a simplified explanation of what a vector database is. Let’s say I come up with the following explanation:

Imagine having a library of books. You can add books to the library 
and retrieve them quickly by using their name or any other indexing method.
A vector database is also a library, but instead of books, stores vectors.
A vector can be thought of, as a list of numbers that represent an image,
audio, or any sort of data. Once we convert the data into vectors, by using
various machine learning techniques, we can store the vectors and retrieve 
them efficiently using some indexing method.

Alternatively, we could ask ChatGPT to provide us with an explanation of the concept if we have trouble writing it ourselves. We can use this prompt to ask ChatGPT:

Explain [concept] to me as if I am a complete beginner with no prior knowledge.

Step 2: Once we have an initial definition of our concept, it’s time to identify how we can improve upon it. We can ask ChatGPT to act as our audience and give us feedback. The prompt for this:

Identify potential gaps or unclear points in my explanation of [concept]: [explanation]

By using this prompt and my explanation of vector database, I get a list of suggestions on which parts of it could have knowledge gaps or more clarification:

Expand on Vectors: Explain how vectors are derived from data, 
using a specific example to illustrate the concept.

Detail Conversion Techniques: Mention specific machine learning techniques,
like neural networks or NLP, used to convert data into vectors.

Clarify Indexing and Retrieval: Discuss specialized algorithms used in 
vector databases, such as approximate nearest neighbor search, and how 
they differ from traditional indexing methods.

Include Use Cases: Add examples of practical applications like multimedia 
retrieval and real-time recommendation systems to show the benefits of 
vector databases.

Discuss Scalability and Performance: Mention how vector databases are 
optimized for handling large datasets and high-performance needs.

Enhance Comparison with Traditional Databases: Extend your library analogy
to compare the search processes in vector and traditional databases, 
emphasizing efficiency and accuracy in finding similar items.

Step 3: Repeat the previous steps until you reach a concise and complete definition of your concept that you are happy with.

Step 4: Expand your explanation by adding analogies and diagrams. Using gpt-4o you can also retrieve diagrams and images to help you get a visual understanding.

I will use two separate prompts, one for analogies:

Add two analogies to help develop a more understandable explanation of [concept].

Another prompt for drawing a diagram of the concept:

draw me diagrams to help understand the concept of [concept]

ChatGPT will continue to create a diagram for a full understanding of vector databases:

The generated diagram by GPT-4o.

☢️WARNING: It is crucial to have in mind that AI hallucinates! This means that it tends to make up information that doesn’t exist. To make matters worse, AI sounds confident in making up these mistakes, so unless you already have some prior knowledge about a topic, fully handing the steering wheel to AI needs caution!

Thanks for reading!

~ Hesam

[1] Richard Feynman Thinking Part 1 of 2

[2] Feynman’s Lost Lecture