How to Use AI to Learn Anything Faster

I want to be careful with this because the framing of "learn anything faster" can make it sound like some kind of cheat code, and it isn't. What AI actually does is remove the friction from certain parts of learning that used to slow me down the most. It doesn't do the learning for you. But it can collapse a lot of the stuff that used to take weeks into a single afternoon.

Here's the experience that made me take this seriously. Last year I wanted to understand how large language models actually work, not surface-level "it predicts the next word" explanations, but mechanistically, what's happening. I had no formal math background. I'd taken one stats class in college and mostly forgotten it.

In the old approach I would've bought a textbook, gotten confused on chapter three, watched some YouTube videos, gotten bored, and given up after two weeks. Instead I used AI as my primary learning tool, and six weeks later I could genuinely follow the arguments in research papers. Not every paper. Not perfectly. But enough to feel like I'd crossed a real threshold.

The thing most people skip: explain your actual level

The number one mistake I see people make when using AI to learn is asking a question without giving context. "Explain transformers" gives you a Wikipedia-quality answer. "I understand basic programming and I've read about neural networks but I can't follow any explanation of attention mechanisms, explain it to me like I understand functions and loops but nothing more advanced" gives you something you can actually use.

I'm not joking when I say this one habit is responsible for 80% of the improvement I've seen in using AI for learning. The more specific you are about exactly where your understanding breaks down, the better the explanation you get. AI doesn't know where you're confused. You have to tell it.

And here's something I do that feels a little weird but works: I tell it when it got the level wrong. "That was too simple, I actually understood that part, it's the next step I'm losing you on." Or "Okay that was too fast, can you slow down on just the part about backpropagation." Having that back-and-forth in real time is something no textbook can do.

Use it to build a map before you go deep

This is one of my favorite techniques and I wish I'd figured it out sooner. Before I start learning anything new, I ask AI to give me the conceptual map of the whole subject. What are the main ideas? What do I need to understand before I can understand the more advanced stuff? What are the common misconceptions beginners have? What are the things experts argue about?

I did this with machine learning and it was genuinely useful. Getting that map first meant I didn't waste time going deep on things that weren't foundational. I knew what to learn first and why. That sounds obvious but when you're learning something new, you often don't know what you don't know, and that map gives you a skeleton to hang everything else on.

The Feynman technique, but with AI as your student

This one changed how I study completely. The Feynman technique is basically: if you can explain a concept to someone who knows nothing, you understand it. If you can't, you don't.

I use AI as my fake student. I'll say "I'm going to explain attention mechanisms to you and I want you to ask me questions wherever my explanation is vague, wrong, or doesn't make sense." Then I explain it. And it pokes holes. And I have to fill the holes.

It's kind of uncomfortable actually, which means it's working. Every time I think I understand something and then AI asks "wait, you said the weights update based on the error, but how does that work when the function isn't differentiable?" and I don't have an answer, I know exactly what I need to go learn next.

This is probably the most effective studying technique I've ever used. More than flashcards, more than note-taking, more than re-reading. Explanation with feedback is just how the brain cements stuff.

Active practice, not passive consumption

Here's where I think a lot of people go wrong with AI-assisted learning: they read explanations and feel like they're learning, but they're actually just consuming. Feeling like you understand something while reading is not the same as understanding it.

I force myself to actively do things with the knowledge. For technical subjects: I ask AI to give me practice problems, and I work through them before looking at solutions. For writing-related skills: I do exercises and then ask for feedback on my specific work, not general tips. For conceptual subjects: I try to apply the concept to something in my real life and then describe my thinking to see if I've got it right.

The difference in retention is massive. If I just read explanations, I forget most of it in a week. If I use it, explain it back, and get corrected on my specific mistakes, I remember it.

Honest about what it's not good for

I could be wrong about some of this, but I've noticed AI-assisted learning has real limitations. Fast feedback works great for conceptual subjects with relatively settled answers. For things that are actively contested in a field, or for skills that require physical practice, or for developing aesthetic judgment, AI is less useful.

I also think there's something to be said for the slow, effortful struggle with hard material that AI can sometimes shortcut in a way that's actually bad for you. I've caught myself asking for an explanation when what I actually needed was to sit with the confusion longer and figure it out. The discomfort of not-understanding is part of how hard learning works. Don't outsource all of it.

And honestly: AI can be confidently wrong, especially for niche or technical subjects. Always cross-reference anything important with a primary source. I've had AI explain something in a way that was subtly off, not totally wrong, but missing an important nuance, and I only caught it because I happened to also read the original paper. That's on me for not verifying. Don't skip verification just because the explanation sounded good.

What this actually looks like in practice

When I start learning something new now, here's the rough sequence. I ask for the conceptual map of the whole subject. I pick the most foundational concept and ask for an explanation at my exact level. I ask follow-up questions until I can explain it back. Then I explain it back and get corrected. Then I do a practice problem or apply it somehow. Then I move to the next concept on the map.

It's not magic. It's still work. But it's incredibly efficient work, and it's actually kind of fun in a way that staring at a textbook never was for me.