🛠️ Statistics tutor prompt

💡 Use Case

• A tutor that never gives the answer first. I paste a problem and my working; it asks where my reasoning went wrong, one question at a time, and only confirms the answer after I have said it.
• Written on 26 February, the morning after five out of ten on the conditional-probability quiz, when I had read three worked solutions and understood all of them and none of my own mistakes.
• It copies Hale Winter, who answers a question with a smaller question. He does it better and only on Tuesdays.
• Used through both modules of 📂 Statistics for Product Decisions, most heavily before the midterm and for the Bayes weeks in October.

How I use it

1. Paste the problem and all of my working, including the crossed-out part.
2. First line of the prompt, always: Do not give the answer. Ask me one question at a time about my reasoning until I find the mistake myself.
3. Answer in sentences, not numbers.
4. When I get there, ask it for one more problem of the same kind. Do that one on paper.
5. Write the mistake as a flashcard in my own words.

⚖️ Pros & Cons

Pros

• It finds the step where I swapped the condition, and makes me say so.
• It is patient at 22:15 on a Tuesday, when Rasmus Kjær is on a tram and Hale is in his cardigan somewhere.
• My own wrong sentences go into the flashcards afterwards.

Cons

• Twice it agreed with a wrong answer I argued for firmly enough. Anything that matters gets checked against the lecture notes.
• It is slower than reading the solution, which is the point and still annoying.
• It will happily let me believe I understand 🔤 p-value. I have a separate note for that.

🔄 Alternatives

• The worked solutions on the course page. Good for checking, bad for learning.
• Petra Novak's office hour, which is better and is on Thursdays at four.