Why ChatGPT Never Makes You Actually Learn Anything

I've asked ChatGPT a few hundred questions.

Not once did I walk away feeling like I had actually learned something.

Getting an answer costs almost nothing now. You type a few words, or hold the voice button and talk, and a reply lands in seconds.

The more particular among us open two more AIs and spend an extra 30 seconds comparing which one gave a better answer, treating answers like Tinder matches.

A few months ago, Google started putting Gemini's summary at the top of the results page, and most people skip clicking into a page at all (including me).

A difficult problem that used to take a week now takes seconds. It leaves almost no trace in your head. No suffer, no learn.

The supply of answers has overshot. But after 50 conversations with AI, strangely, I don't feel like I know more. I have more questions, and I'm very tired.

The reward sits in the wrong place

I sat with this for a while. Tried to approach this question with cognitive science. Here's the conclusion I came to: the quality of the answers has nothing to do with it. It's the loop the interface is built around.

The chat-based interface everyone is used to rewards one action: asking the next question.

You hit send, the reply comes back, dopamine pumps up.

The reward sits in the wrong place, and it pulls you into asking again. You can fire off dozens of questions without ever stopping, answers the whole way down.

Asking without stopping can hand you a fake sense of progress because understanding does not happen there. It happens when you stop, you reflect, you chew on the problem, you think about the problem so hard until everything looks like an inspiration of the solution.

Three things that decide whether you learn

I studied cognitive science, the field that looks at how people remember, think and learn. Three things matter if you want to actually learn something.

• Cognitive load: Learning runs on working memory. Working memory holds very little at once, and the ceiling drops as you get older; once it's over the line, nothing else goes in. What matters is the split inside it. The difficulty of the material itself should stay, that's the thing you came to learn. Messy layout, jargon dropped in with no setup, jumping around, those are pure burn on your budget. The budget you get back is what pays for the real work of building long-term memory.
• Retrieval practice (quiz): The circuits in your head grow while you're recalling. When you recall, you have to walk the whole logic yourself, and that's when information sets into a structure. Taking information in is like carrying planks from the yard into the house. Retrieval is nailing the planks into a cabinet. A room full of planks doesn't turn into a cabinet on its own.
• Desirable difficulties: Bjork's term. Some difficulty makes you feel like you're learning badly in the moment while it actually sticks better, and your brain keeps working on it while you aren't thinking about it. When you hit a hard problem, chew on it for a while before you check the answer. Same as lifting dumbbells, the resistance is the training. Picture yourself training for weightlifting with a robot next to you lifting three hundred kilos on your behalf. The muscle still doesn't end up on you.

Hold those three up against the chat interface and it's clear why nothing feels learned. Inside that question-and-answer loop, AI slop piles on extra cognitive load, there's almost no point where you had to stop to recall and organize, and desirable difficulty goes down to zero. The whole design of the chat-based interface runs against decades of research on the brain.

So, here's how you actually learn

The core: immersive reading

Yes, unfortunately, you still have to read in the AI era. And not just read — you must read without interruption, without hesitation, and without the thought of "this is too hard" lingering in your mind. You need a mindset of genuine curiosity: "What is this? I want to figure it out."

Block out 20 to 30 minutes for a single session. Focused attention is all you need. Completing a session provides an intrinsic reward, forming a positive reinforcement loop that strengthens the habit for next time.

The supporting tactics

• Calibrate the difficulty: Ensure the material is at the correct level of cognitive load. It shouldn't be too hard or too easy. You are aiming for that "desirable difficulty" sweet spot — challenging, but manageable.
• Structured format: A messy layout is a massive source of unwanted cognitive load. The format must be clean, structured, and concise. Highlight key concepts and place footnotes strategically, so you don't have to hold unfamiliar terms in your working memory until the end of the chapter.
• End-of-chapter quizzes: This is to wrap up, review, and most importantly, prompt active recall. Recalling is like picking up items scattered across the floor and organizing them onto a shelf. You categorize them, color-code them, and make sure everything is exactly where it belongs. Memory consolidation happens every time you force your brain to retrieve information.

To put these principles into practice, I built a studying tool designed entirely around the last 20 years of cognitive and brain research. You give it whatever you need to read; it comes back with calibrated chapters, a clean structure, and a quiz at the end of each one.