AI study help: use hints after trying the question
Try the question first, ask for a focused hint and continue the working yourself. Check your understanding on a new question without assistance.
A 24/7 AI tutor that always has the answer is a homework solver with friendlier copy. The 2025 PNAS field experiment is the caution: a ChatGPT-like tutor raised practice-with-help and cut scores when the tool was removed. A “GPT Tutor” that gave hints and withheld full solutions largely avoided that harm. RubricMark is not that paper. We take the design implication: hints after attempt, not unrestricted answer chat as the study UX.
Learn more: AI tutor — hints after you try.
The one-hint protocol
- Write a full attempt or a stuck-point (“I have the rate; I don’t know which quantity is the base”).
- Ask for one next step or a question back — not the finished method.
- Produce the next line yourself.
- Only then check. If you still miss, log it in the Mistake Book and retrieve tomorrow.
If the tutor writes the solution in one shot, you just read. Fluency is not mastery — that is the illusion of competence the active-learning pillar keeps naming.
Where Rubi is allowed to tutor
- After a sit — why this MCQ was wrong; which criterion in a writing report; what to drill next.
- After you write — assignment feedback with quote-anchored marks. Not “write my essay.”
- Around cards — draft a deck from notes; you still flip. Improve a card you review before save.
- Never on the live attempt path — Practice Hub and Career Hub sample exam-style questions. No live-invented MCQs, no in-runner “just tell me.”
That is AI as enhancer: it diagnoses and queues. You still retrieve. Success is the next unassisted score, not how fast the chat finished homework.
If you already use a general chatbot as a tutor
Keep it off the first attempt. Paste your stuck-point, not the blank stem. Ask it to withhold the answer. Then close the tab and sit a practice test for your exam in Practice Hub — JLPT, HAST, IB, IELTS, GAMSAT — so you find out whether the hint transferred. Australia’s school AI framework is the same idea at policy level: purpose-built help, not shadow paste-into-consumer-chat for every worksheet.
Previous: word problems. Next: AI quiz from notes vs a timed mock.