How to study with AI without skipping the learning

You can finish an exercise with AI assistance and still be unable to solve it yourself. When studying with AI, make your own attempt, ask about the step that stopped you and finish with a fresh question without help. That check tells you more than counting generated answers.

An interview with psychologist Carlos Gantiva in La Silla Vacía, found in Publer’s technology feed on September 12, 2026, discusses screens, attention and learning. An interview alone does not establish that every use of AI reduces learning ability. We use the discussion as a starting point for a practical study routine.

Bastani and colleagues tested GPT-4 tutors in secondary mathematics in Turkey. Unrestricted assistance improved practice performance but worsened a subsequent unaided assessment relative to the control group. A tutor with teacher-informed hints mitigated that effect. The experiment did not test Lumia AI or establish permanent harm or universal outcomes across subjects.

Our practical suggestion is to check assisted work separately from what you can do independently. A clear summary can help you get oriented. To assess understanding, produce an explanation or solve a problem. Keep both attempts and compare them with a reference you can verify.

Suppose a backpack costs 80 and is discounted by 25%. Write your calculation before opening a chat. If you cannot begin, ask a specific question: ‘I do not understand how to turn a percentage into an amount. Give me a hint about that step.’ Keep your attempt even when it is wrong.

Then check the procedure against your notes: 25% of 80 is 20, so the final price is 60. Explain why you subtract 20 rather than 25. Next, change the numbers: an item costs 120 and has a 15% discount. Solve it without consulting the conversation. The discount is 18 and the final price is 102. If you can only repeat the first example, revisit the concept before moving on.

Adapt these instructions to an AI assistant. They are suggested prompts, not a guarantee that the model will follow every condition. If it reveals the solution too soon, use another exercise and attempt it without looking. Verify hints too: a convincing explanation can contain mistakes.

Use Lumia AI to prepare draft flashcards from your study material. Check every question and answer against the original source before reviewing. Each card should test a clear idea. ‘How do I calculate the discount?’ and ‘How do I find the final price?’ are easier to assess separately than in a long answer containing several steps.

During review, answer before revealing the solution and assess your difficulty honestly. Then complete an exercise outside the cards to check application. The related guides to quiz mistakes and source-based study expand on checking your practice. This is a suggested routine; we do not claim that Lumia reproduces the research tutor or its results.

Close the material and write a short explanation. Record one unanswered question and one mistake you corrected. In another session, attempt a different question before rereading. Choose what to review based on your answers and the steps you can explain.

When an attempt fails, decide whether you need an earlier explanation, a clearer question or problems with different numbers. Change one thing at a time. Use AI to help prepare and check practice while keeping real opportunities to think, make mistakes and assess what you have learned.

If you use AI to prepare a quiz, check its answer key before interpreting your results. An ambiguous question or false explanation needs a correction in the material; your own confusion calls for further practice. Keep the attempt and describe the error precisely: definition, condition, procedure, or question interpretation.

The related guide to reviewing quiz mistakes shows how to turn an error into a card or a new problem. The next session should check the correction without relying on the same options. Keep the record in your notes; do not assume the assistant can identify the cause of every error automatically.