The best way to learn from AI is to really reverse the whole process like to use ai for deep understanding rather than just "getting the answer" revert the typical workflow, instead of asking the AI to explain the text to you (passive) you should use the AI to force you to explain the text (active).
This somewhat prevents the "Illusion of Competence" the false feeling of mastery that comes from reading a clear summary without doing the cognitive work to create it. Like its kinda the more the AI does the less you do and the less you remember
So the only solution for this which I use on a daily basis is to not use ai as a tutor but someone who questions your understanding and someone who evaluates your explanations. So the question is how do we do it ??
Executing it
So whatever I will mentioning here will be using something called active recall principle and overall it changes from something that only utters and gives you answers to something that questions your understanding.
Step 1 : Preparation | The chunking part
The goal for this step is to break the PDF into manageable cognitive loads without summarizing it yet and my suggestion would be to not upload the entire textbook and ask for a summary this will encourage more passive skimming. Instead upload the PDF and use a prompt that forces you to identify the structure first
Prompt :
"I am uploading a lecture/chapter on [Topic]. Do not summarize it. Instead, scan the document and list the top 5 'Core Concepts' and the 3 'Most Complex Arguments' presented. For each, provide only the heading and a reference to the page number. I will then study these sections myself."
this works for me because it forces me to build a mental model to unfold topics before I start filling in details
Step 2 : The Deep Work
Once you are done with a section yourself choose one of these two modes to test your understanding.
Mode A is where you teach ai Kinda very helpful method for retention where you attempt to explain the concept to the AI and it critiques you. This is also known as the famous Feynman Technique where you learn or read something and try to explain those learnings to a 5 year old.
Prompt :
"I have just studied [Concept X] from the uploaded PDF. I am going to explain it to you as if you were a 12-year-old student. Your Role: Listen to my explanation. If I miss a key nuance, get a fact wrong, or use jargon without defining it, stop me and ask a clarifying question. Do not just correct me ask me to clarify. Grade my explanation on clarity and accuracy at the end."
This works for me while I am reading some white papers related to Backend infrastructure and complex architecture of a component the teaching helps me with my retention of these architectures for a longer period of time.
Mode B is all about AI teaching you This is mostly used by me when I am stuck or need to test my logic it happens while reading white papers/blogs/articles related to diffusion models, LLM math, AI infra.
"Act as a strict professor testing my understanding of [Section Y] in the PDF.
Rules:
- Ask me one question at a time.
- Do not give me the answer. If I am wrong, give me a subtle hint or ask a follow-up question that highlights my error.
- Start with a broad conceptual question, then drill down into the specifics."
It forces me to not have habit of passively reading an answer and make a habit of retrieval practice instead.
For the mode A you can take in more references such as Protégé Effect which helps the teacher retention and boosts it for the subject he/she teaches
- https://thelearnerlab.com/protege-effect/
- https://www.psychologytoday.com/us/blog/how-to-be-brilliant/201206/the-protege-effect
Step 3 : The Testing phase
In my experience most ai gen quizzes are way too easy to just pass. To make them more effective for memorizing stuff I ask for more and more wrong answers that look right
Prompt :
"Create a 5-question multiple-choice quiz based *strictly* on the evidence in the PDF.
**Critical Instructions:
- **Make the questions 'Application-based' (e.g., 'Which scenario best applies the theory?') rather than definition-based.
- ensure the wrong answers (distractors) are plausible they should represent common misconceptions or partial truths mentioned in the text.
- Do not reveal the answers until I respond.
Identifying why the answer is wrong helped me build more on my critical thinking than spotting an obviously wrong one and helped me diving deeper into the few concepts.
Step 4 : Synthesis
This one just makes sure you are seeing the big picture and if not just ask it to help you connect the concepts.
Prompt :
"I have studied concepts A, B, and C from the text. I think the relationship between them is [Insert Your Theory: e.g., 'A causes B, which leads to C']. Is this accurate based on the text? Where does my logic break down?"
A quick disclaimer all these prompts are refined with the help of GPT and Kimi K2 but the context behind using these prompts stays as it is
