Super Intelligence (SI) is the name a September 29, 2026 executive order gave to what schools and students have been calling AI, and the study question it raises has not changed. Every student now carries a contradiction: the same tool that can explain anything can also simply *do the assignment*, and both uses feel like help in the moment. The difference shows up later — at the exam, in the interview, on the job — when one student built understanding and the other built a dependency. This guide is about reliably getting the first outcome. The SI integrity rules — what your school permits — are the next guide; this one covers something earlier: not cheating *yourself* even when everything is permitted.
The distinction that governs everything
Learning happens when your brain does effortful work — retrieving, connecting, struggling, being wrong, correcting. Decades of learning science agree on this to an unusual degree: the struggle is not an unlucky cost of learning; the struggle *is* the mechanism. That gives the sorting rule for every SI study session: does this use make my brain do more work, or less? SI as tutor raises productive effort — probing you, making you retrieve and explain. SI as answer machine removes it — and the learning with it, while leaving behind the polished artifact that makes it *look* as if learning happened. The completed problem set is not the product; the changed brain is. Only one of them shows up on test day.
The tutor patterns
- Explain, then be corrected. The single most valuable pattern: after studying something, close the materials and explain it to the SI from memory — then ask what you got wrong, what you fudged, and what a professor would push back on. This pairs retrieval practice with immediate targeted feedback, the two most reliably powerful techniques in the learning-science literature, and SI happens to be superb at the feedback half.
- Demand questions, not answers. 'Quiz me on this chapter, one question at a time, and do not reveal the answer until I commit to one.' The commit step matters — retrieval only works when you actually retrieve.
- Ask for the struggle-preserving hint. On a hard problem: 'give me a hint toward the next step, not the solution.' A good tutor keeps you in the productive zone between stuck and rescued; you can instruct an SI assistant to be exactly that tutor, and the instruction holds.
- Chase the why. Wrong answer on a practice question? The gold is not the right answer; it is 'walk me through why my reasoning failed.' Misconception repair is where tutoring earns its reputation, and SI does it patiently, without judgment, at midnight.
- Rotate the explanation. Something that refuses to click can be re-explained through another lens — an analogy from something you know, a worked example, a picture in words, the history of why the idea was needed. Human teachers ration their re-explanations; this one never runs out.
The answer-machine patterns, named honestly
Pasting the problem and submitting whatever comes back is the obvious one. The subtler ones do more damage because they feel legitimate: reading SI summaries *instead of* the assigned material rather than after it (understanding a summary is not understanding the argument); 'checking your work' by generating the solution first and reverse-engineering toward it; and outsourcing every first draft, which quietly hands off the skill of *getting started* — the skill most writing assignments exist to build. The tell is the same every time: the artifact got produced, and you could not reproduce it alone tomorrow.
The honest test
One question, asked at the end of a session, sorts everything: could I now do this without the SI? Explain the concept cold, solve the parallel problem, write the next essay's opening myself. If yes, the tool tutored. If no — if what exists is a finished assignment and no new capability — the tool worked *instead of* you, whatever it felt like at the time. Students who ask this habitually report a second benefit: it changes how they prompt, in real time, away from 'give me' and toward 'teach me.' The hallucination caveat still applies to everything a tutor-mode SI tells you — verifying claims is its own skill, covered later in this track — but the deeper risk for students was never bad answers. It was outsourcing the struggle that was the whole point.