How to Use AI to Learn Anything Faster (A Beginner's Guide)

The Feynman technique is a classic learning method developed by the physicist Richard Feynman. The idea: you don't really understand something until you can explain it in plain language to someone who knows nothing about it. The act of trying to explain forces you to confront the bits you don't actually understand.

The traditional version of the Feynman technique requires a tame friend or family member willing to listen patiently to you explaining quantum entanglement at the dinner table. AI gives you an infinitely patient version who can also push back intelligently.

The workflow:

1. Pick a topic you want to understand - anything from "how a mortgage works" to "what cell signalling pathways do" to "how the Tudors lost Calais."
2. Open ChatGPT (or your chat tool of choice) and say: "I'm trying to learn [topic]. I'm going to explain it back to you in my own words. Please point out any factual errors, any places where my explanation is confused or oversimplified, and anything important that I missed. Don't be too gentle - I want corrections, not encouragement."
3. Write out your understanding of the topic in 3-5 paragraphs. Don't look anything up while you're writing.
4. Submit it. Read the AI's response carefully - the corrections are the gold.
5. Repeat for any sub-topic where the AI flagged confusion.

The AI version of Feynman has one advantage over the original: the chatbot won't get bored or politely change the subject when your explanation gets bad. Use that. The willingness to be told you are wrong is the whole game.

A prompt template that works well: "Critique my explanation of [topic] as if you were marking an A-level (or undergraduate, or postgraduate - adjust to the level you are working at) answer. Highlight: (a) factual errors, (b) conceptual confusions, (c) important things I left out, (d) bits where I oversimplified to the point of being wrong. Be specific about what to add or change."

ELI5 - "explain it like I'm 5" - is a Reddit-coined phrase for asking for a very simple explanation of a complex topic. The technique exploits one of the underrated things AI chat tools are good at: gracefully adjusting the level of explanation to the audience.

The trick is to start at the ELI5 level even if you are a competent adult, and then deliberately ladder up. A sequence that works on basically any technical topic:

1. "Explain [topic] as if I'm 5 years old. Use everyday objects and short sentences. No technical terms."
2. "Now explain it as if I'm 12 years old. You can use one or two technical terms if you define them."
3. "Now explain it as if I'm a first-year undergraduate in a related field. You can assume basic background but explain anything specialist."
4. "Now explain it at the level of a textbook chapter for someone studying this field as their main subject."
5. "Now write the version that would appear in a current research paper introduction."

Each step builds on the previous one. By step 3 or 4 you usually have a mental model that lets you read primary sources without bouncing off the jargon. The ELI5 starting point gives you the metaphor that everything else can attach to.

This technique is particularly good for topics where the technical literature is gatekept by terminology - economics, biology, statistics, machine learning, law. The ELI5 version isn't a permanent simplification; it's a temporary handle you grab while you build the more careful understanding on top of it.

Active recall - testing yourself on material rather than passively re-reading it - is the most evidence-supported learning technique in the cognitive psychology literature. It is also boring to do alone, which is why most people skip it. AI removes that friction.

The workflow:

1. Read or skim a piece of material - a textbook chapter, a course handout, a research paper, your own notes.
2. Paste it (or describe it) to the AI and say: "Generate 10 short questions that would test whether someone has properly understood this material. Mix factual recall (about a quarter of the questions), application questions (about half - 'how would you apply this in scenario X'), and conceptual questions ('explain the relationship between A and B in your own words'). Don't give me the answers yet - I want to attempt them first."
3. Answer the questions in writing without looking at the source.
4. Paste your answers back: "Here are my attempts. Mark them. Be specific about what I got right, what I got wrong, what was incomplete, and which gaps in my understanding the answers reveal."
5. Re-read the source for any topics where the AI flagged gaps.

This is more demanding than passive reading and considerably more effective. Self-quizzing tends to produce far more durable retention than passive re-reading - the 'testing effect' is one of the most replicated findings in learning research. AI makes the self-quiz part essentially free.

A refinement worth knowing: after you've done this once, ask for a spaced repetition of the questions you got wrong. "In one week, send me back the three questions I got wrong, plus two new ones on the same topics - I want to test whether the corrections stuck." The chatbot won't actually message you a week later, but if you save the conversation and re-open it then, the structured re-test works.

Most adult learning is bottlenecked by reading time, not comprehension. A 400-page textbook, a 30-page research paper, a court judgment, a thick technical specification - the volume of material is the obstacle. AI is unusually good at summarising long documents, which lets you read selectively.

The key insight: do not use AI summaries as a substitute for reading the source. Use them as a map that tells you which 20% of the source actually deserves your attention.

The workflow:

1. Paste (or describe) the source material to the AI.
2. Say: "Summarise this in three layers. Layer one: a one-paragraph summary of the whole document - what's the core argument or finding. Layer two: a one-sentence summary of each major section or chapter - show me the skeleton. Layer three: the three things this document says that a knowledgeable reader would find genuinely surprising or non-obvious."
3. Read the layered summary. Use it to decide which sections of the original to read in full.
4. Read those sections properly, in the original. Skip or skim the rest.

This is particularly effective for textbooks (most chapters can be skim-summarised; the 1-2 chapters that contain the load-bearing concepts need careful reading), academic papers (the methods and key results sections almost always reward close reading; the introduction and discussion can be summarised), and any reference material where you want to know what's there without consuming all of it.

A related variant: "What is this document's strongest claim, and what is the weakest part of its argument for that claim?" AI is unusually good at this kind of structural critique - it's the same skill as the Feynman-technique critique, applied to existing writing instead of yours.

Imagine you have to give a 10-minute talk at work on "how DNS works" by next Friday, and you currently know about as much about DNS as the average person (i.e. "it's the internet's phone book - beyond that, no idea"). The five-technique combined workflow looks like:

Day 1 (1 hour total): Open ChatGPT. Run Technique 2 (ELI5 to undergraduate-level laddered explanations of DNS - about 20 minutes). Then run Technique 3 (build a 4-day study plan given that you have 3-4 hours per day available, you understand the ELI5 version, you don't understand how DNS records actually propagate or what root name servers do - 15 minutes). Then close the laptop.

Day 2: Read whatever your AI-generated plan suggested for day 2 - typically a beginner-friendly tutorial article and a short technical primer. Then run Technique 1 (Feynman-explain DNS resolution back to ChatGPT in your own words; have it correct you - 30 minutes).

Day 3: Read the deeper material the plan suggested (typically a Cloudflare or AWS technical primer plus the relevant Wikipedia article). Then run Technique 4 (self-quiz on what you've read - 30 minutes).

Day 4: Outline the talk. Use Technique 5 to summarise any reference material you want to draw quotes or stats from. Run Technique 1 once more, this time on your draft talk: "Critique this talk script. What's confusing? What did I overclaim? What did I miss that the audience will ask about?"

Friday: Give the talk. You'll probably do a reasonable job.

The gain over the alternative - "Google around for 4 days reading random tutorials" - is mostly in the structure and feedback. The AI replaces the bored friend who would otherwise be giving you the feedback. The depth of your final understanding depends on how much real material you read in the middle steps - AI does not bypass the reading, it makes the reading targeted.

Three categories of learning where AI is less helpful than the marketing implies:

Skills that require physical practice. AI can explain how to do a backhand drive in pickleball, how to weld, or how to play the chord changes in a jazz standard. None of those explanations build the muscle memory. For physical skills, the AI tutor speeds up the conceptual understanding; the practice still has to happen.

Cutting-edge specialist knowledge. AI models are trained on data up to a cut-off date and may not reflect very recent developments - last year's regulations, last month's research paper, this week's company announcement. For topics where being current matters, AI is a starting point, not an authority. Always cross-check recent claims against primary sources.

Cases where being wrong is expensive. Medical decisions, legal questions, regulatory compliance, financial-product specifics, anything where one wrong number could materially harm you. AI is a teaching assistant in these areas; the authority needs to be a textbook, a regulator's published guidance, or a qualified human professional.

None of these means AI is useless for those topics - it means the verification step matters more. The Feynman-technique critique and the self-quiz both have the side benefit of surfacing claims you should double-check before relying on.