There is a specific kind of suffering that hits at 1 a.m. when you've been staring at a calculus problem for forty-five minutes, the textbook example uses different notation, the lecture slides skip the step that's confusing you, and your professor's office hours are in three days. The shame of not getting it makes you not want to ask anyone. You close the laptop and decide you're "not a math person."

You are a math person. You just didn't have a tutor available at 1 a.m. Now you do.

Tutor mode, not answer mode

The single most important thing in this chapter: stop asking AI to give you answers. Ask it to teach you.

There is an enormous difference between these two prompts.

Solve this integral: ∫(3x² + 2x)dx

I'm trying to learn how to solve this integral: ∫(3x² + 2x)dx. Walk me
through the thought process step by step. After each step, ask me what
the next move should be before you do it. If I get it wrong, explain
why and let me try again. Don't give me the full answer until I've
worked through it.

The first prompt gives you a number. The second prompt gives you the ability to solve the next twenty integrals on your own. Same tool. Wildly different outcomes.

Use a saved version of this template for everything: math, physics, engineering, statistics, programming, economics, anywhere there's a problem to work through.

Tutor me through [problem]. Use Socratic method — ask leading questions
instead of giving answers. Wait for my response before continuing.
Correct mistakes patiently. Explain the underlying concept, not just
the mechanics.

The hallucination problem (especially in math)

AI gets math wrong. Not occasionally — often, especially with multi-step arithmetic, large numbers, or anything requiring exact precision. ChatGPT can confidently tell you that 17 × 23 is 381 (it's 391). It can drop a negative sign in the middle of an algebra step and ride that error all the way to a wrong answer that still looks plausible.

Three rules for using AI on math:

1. Always verify the final answer. Plug it back into the original equation. If it doesn't work, you have a bug to find together.
2. Watch for sign errors and arithmetic slips in the middle steps. They are extremely common. Read every line.
3. For anything beyond simple algebra, use the right tool. ChatGPT and Claude can both run Python under the hood (Code Interpreter / Analysis tool). Turn it on. The tool will compute exactly instead of guessing. Or use Wolfram Alpha for raw computation.

A good prompt acknowledges this:

Solve this step by step. Before stating the final answer, verify it by
substituting back into the original equation. If you used arithmetic
in any step, redo it twice and confirm. Use the code tool if available.

Tool selection by subject

Calculus, linear algebra, differential equations. ChatGPT or Claude with code execution turned on, or Wolfram Alpha for raw answers. For exact computation, force the code tool.

Proofs and pure math. Claude is noticeably better at constructing and critiquing proofs. It tends to acknowledge when it's unsure rather than confidently inventing a step.

Physics. Either chatbot, with code on for numerics. For diagrams, Gemini or ChatGPT (both handle image input). Photograph the diagram, upload it, ask for a setup walkthrough.

Organic chemistry. Claude or ChatGPT for mechanism explanations. Verify mechanisms against your textbook for high-stakes work. For the midnight "why does this electron go here" question, they're infinitely better than nothing.

Explain the SN2 mechanism for this reaction step by step. For each
arrow, tell me which atom is attacking which, why that bond forms, and
why the leaving group leaves. Treat me like I just learned what an
orbital is.

Statistics. Force the code tool. Ask AI to write Python that does the actual computation, then explain what each line does. You'll learn stats and Python at the same time.

Programming. See Chapter 8. Short version: Cursor is the IDE that puts AI directly into your editor. Cursor AI IDE Workflows covers the setup.

Slides and lectures. NotebookLM. Upload the deck. "Explain slide 14, I don't understand the equation."

Catching wrong answers

You are going to be wrong sometimes. AI is going to be wrong sometimes. The skill is figuring out who is wrong.

When AI's answer disagrees with the back of the textbook or with your own work, do this:

The textbook says the answer is X. You said Y. Without looking up the
correct answer, walk through both solutions and tell me where they
diverge. Identify the exact step where one of us makes a different
choice and explain what's happening at that step.

About 60% of the time, AI made an error and now you can both see it. About 30% of the time, you made an error and the diagnosis reveals it. About 10% of the time the textbook is using a different convention or has a typo. All three of those outcomes teach you something.

When AI just isn't getting it

Sometimes you'll be stuck on a concept and AI keeps explaining it the same wrong way. Don't grind. Try these escape hatches:

• Switch models. Take the same prompt to a different chatbot. Different training, sometimes a clearer angle.
• Ask for a different metaphor. "Explain it again, but use an analogy from sports/music/cooking."
• Ask what's missing. "I'm clearly not getting this. What background knowledge am I probably missing that's making this confusing?"
• Read the actual textbook. Old technology. Still works.

A subject-specific cheat sheet

Calculus: "Walk me through this derivative/integral step by step, ask me to predict the next move at each step, and verify the answer by [differentiating back / plugging in]."

Organic chem: "Explain this reaction mechanism arrow by arrow. After each arrow, ask me what should happen next. Reference the specific orbitals involved."

Statistics: "Walk me through choosing the right test for [scenario], computing it (use code), and interpreting the result for a non-statistician."

Proofs: "Before writing any of the proof, walk me through the proof strategy — what do I need to show, what techniques are likely to work? Then we write."

The students who stop being afraid of hard subjects don't get smarter. They just get better tutors. You now have one in your pocket, on call, infinitely patient. Use it.