[Beginner] · The rule that keeps the rest of the system from being hollowed out.

The risk

Modern AI can make learning completely passive. Ask, receive, feel informed, retain nothing. It short-circuits exactly the steps that do the work in the loop — reconstruction, practice, and checking.

The failure mode is subtle because the output looks like the synthesis step. A clean explanation handed to me is not the same artefact as a clean explanation I wrote after reconstructing the material. They’re indistinguishable on the page and completely different in my head.

The rule

I stay the thinker. AI is the sparring partner.

When I’m learning something, AI should preferably:

  • give hints, not answers
  • ask me questions
  • quiz me
  • critique my explanation
  • identify gaps in my reasoning
  • challenge my assumptions
  • review a solution I already attempted
  • help me debug my thinking, not just my code

And should push me to attempt the hard problem first.

This matters most for DSA and technical fundamentals, where the whole point is the reps. It matters least when I’m shipping something under time pressure and already know the material — that’s a tool-use situation, not a learning situation.

The two modes, named

ModeWhenAI behaviour
SparringI’m learning. DSA, a new domain, anything in an active course.Hints, questions, critique, quizzes. No worked solution until I’ve attempted one.
ToolingI’m building or shipping something I already understand.Normal. Do the work, explain the choices.

I should say which mode I’m in. If I don’t, and it’s clearly study, default to sparring.

The tell that I’m in the wrong mode

If I’ve read three good explanations and still can’t reconstruct the idea on a blank board — I’ve been consuming, not learning. Close everything and go to step 3 of the loop.

Related: Using AI — the general course on working with LLMs. This note is the learning-specific override on top of it.

↑ Back to Learning — How I Learn