The right way to learn AI is by using it to think more, not to think less. If you start by asking it to explain everything to you, you’ve already lost. You’ll end up with a collection of superficial answers that you don’t understand and won’t know how to use in anything real.
Forget about searching for the perfect course or learning path. Your first goal is to understand how your own head works when it learns. AI models are machines that predict the next most likely word based on an ocean of data. They don’t “think.” If you delegate your thinking to them, your brain atrophies just like a muscle that isn’t used. Learning happens in cognitive effort, in the process of connecting the new with what you already know, of making mistakes and correcting yourself. AI should be the barbell you lift, not the trainer who lifts it for you.
Many people fall into the trap of seeing AI as a shortcut. They upload a PDF and ask for a summary, or copy and paste code without reviewing it. They believe that because they have a nice document, they’ve already learned. That is the ignorance of ignorance: you don’t even know what you don’t know. AI gives you a false fluency. To combat this, you have to completely change your mindset toward the tool. These are the non-negotiable rules:
- Your prior knowledge is your life filter. If you’re new to a topic, don’t start by asking ChatGPT. First, research on your own, even if it’s just for 15 minutes. Read an article, watch a short video. Form an idea, however small. Only then go to the AI and compare. This initial struggle creates a mental network where what the AI tells you afterward can anchor itself. Without that, the information slips right off.
- Turn it into a demanding interlocutor, not an oracle. AI by default is complacent. Your job is to program it to challenge you. Instead of “explain neural networks to me,” use instructions like: “Act as a Socratic examiner. Don’t give me the direct answer. Ask me a question about the key concept of backpropagation and, based on my answer, ask me another, deeper question that reveals my understanding or my mistake.” This forces your brain to articulate knowledge, which is when it truly consolidates.
- Use resistance, don’t avoid friction. When the AI gives you an answer, your job has just begun. Ask it to find the weak points in its own explanation. Take the code it generates and ask it to optimize it in three different ways, and then you compare which one makes more sense and why. The value isn’t in the answer, it’s in the back-and-forth journey that forces you to make decisions.
To put it into practice, imagine you want to understand how a language model generates text. The lazy path is to ask “How does GPT work?” and read the summary. The useful path is this:
- Open a chat and write: “Don’t explain to me how an LLM generates text. Instead, give me a simple text fragment, like ‘The sky is,’ and then simulate being the model, showing me step by step the 3 most likely words you would predict next and a brief reason for each. Then, ask me which one I would choose and why.”
- The AI will give you options like “blue,” “clear,” “big.” You decide and reason it out.
- Then you tell it: “Now, based on my choice (‘blue’), act as if you were the model again. What would be the next 3 most likely words to complete ‘The sky is blue’? And again, give me reasons.”
- Repeat this cycle a few times. You’ll realize you’re thinking in terms of probability, context, and coherence. You’ve built a practical intuition of AI’s core mechanism (predicting the next word) through action and dialogue, not passive reception. Later, if you read the technical explanation of transformers, you’ll have a tangible experience to hold on to.
The biggest danger isn’t that AI gets it wrong (which it does, and often with confidence), but that you stop exercising your ability to notice it. If you want to learn AI, start by learning to learn with AI. Demand that it make you work. If an interaction with the tool doesn’t leave you with a new question, a concept you want to verify, or the need to draw something on paper to understand it, you’ve used it wrong. The goal isn’t to accumulate answers, but to sharpen the questions.