Code
Turning AI Drafts into Actual Learning
How I use lecture notes, questions, retrieval, and revision to make AI-assisted writing part of studying
I revised a recent post on linear regression several times. The material I gave the AI included keywords and questions from the lecture, the central points from my notes, and explanations I had already written in my own words. Even with that material in the draft, I could not tell whether I really understood the section on polynomial regression.
I kept asking what separated multiple linear regression from polynomial regression, why counts as a second-degree term, and whether the model was learning an increase in an apartment price. When an explanation did not work for me, I changed the example and rewrote the paragraph. The post became longer, but the concepts finally became clearer.
That left me with a more important question. Does revising a draft written by AI count as studying, or can a polished explanation make me believe I understand more than I do?
Writing from memory is different from reading a draft
Writing on a blank page after a lecture forces me to retrieve the material from memory. I have to decide what to explain first, connect ideas that appeared separately, and turn what I understood into my own sentences.
In cognitive psychology, the generation effect describes how producing an answer can make it more memorable than simply reading it. Retrieval practice works through a related action: making an effort to pull information back out of memory. That effort reuses the connections around the memory, which makes the same information easier to retrieve later. The work that makes direct writing difficult is also part of what makes it useful for learning.
Reading a completed draft creates a different experience. A well-organized explanation can feel familiar and easy to follow. Processing fluency is one reason that ease can be mistaken for understanding. I may recognize every sentence while reading it, then fail to explain the same idea after closing the page.
Using an AI draft does not automatically remove learning from the process. What matters is the thinking I do before and after the draft appears.
| Approach | What I do | What I may miss |
|---|---|---|
| Write directly | Retrieve, organize, explain, and phrase the idea | It takes time, and the writing itself can become tiring |
| Read an AI result | Follow a completed explanation | Easy reading can be mistaken for understanding |
| Revise with AI | Select questions, inspect structure, correct errors, and rewrite | Passive reading takes over if I accept the draft unchanged |
Where the learning happens in my workflow
I do not hand over the lecture text by itself and wait for a finished post. I extract the main points, mark the parts I do not understand, and add my own current explanation. That preparation already forces me to separate what I know from what I cannot yet explain.
The AI turns those materials into a draft. I then look for gaps in the reasoning. If an explanation still does not make sense, I ask another question. If an example does not fit, I replace it. I also remove phrases I would never use. The polynomial regression post went through this loop several times.
Collect notes and questions → let AI structure a draft → find what still does not make sense → question and verify it → rewrite it in my own words → commit
The AI handles connections between sentences, initial structure, and formatting. I remain responsible for choosing the important points, noticing what I do not understand, checking the technical explanation, and approving the final wording.
This is not identical to writing everything from a blank page. Because the structure and wording arrive early, I retrieve less of the material on my own. The trade-off is that I can use limited study time to test more questions and spend more effort on the parts that remain unclear.
I do not think I need to choose only one method. For a difficult concept, I can write three or four sentences before asking for a draft. After the post is finished, I can close the screen and explain the concept again. That returns part of the time saved on phrasing to retrieval practice.
Revising an old post can become review
Even a post that was clear when I wrote it will fade from memory. When a later lecture returns to the same concept, opening the old post and extending it creates a chance to compare the earlier explanation with the new material.
Opening an old post is not enough by itself. Spaced repetition means revisiting the same material after time has passed. If I only reread familiar sentences and polish the wording, I may still be following recognition rather than retrieving the idea. Review becomes more useful when I first recall what I can, compare that attempt with the old explanation, and then correct the errors or missing connections.
That is what happened with the regression post. I first added a definition of polynomial regression. On the next pass, I compared it with multiple linear regression. Then I checked the degree of an interaction term and added an apartment-price example. The Git history records changes to the prose, but it also records where my understanding changed.
The schedule mattered. Classes and assignments arrived every day, and I was also trying to publish a post each day. Writing one perfect explanation from the beginning was not realistic. It worked better to write what I understood at the time and return later. Revising also created more opportunities to connect old and new knowledge than starting a separate post for every lesson.
Two rules I want to keep
The use of AI does not decide whether learning happened. If I delegate the questions and the central reasoning, then only read the result, the learning is shallow. If I form the questions first, test the draft, and keep working on the parts I cannot explain, AI becomes a useful study tool.
I want to keep two rules: retrieve what I know before asking for a draft, and recall what I remember before revising an old post. AI can structure the writing quickly, but the questions and the responsibility for understanding remain mine.