Subject deep dive · Math

Open · trust, but verify

Do Poll: "Would you trust an AI's answer to a multi-step arithmetic problem?" Then show, live, that it can be confidently wrong. Let the surprise land.

Focus · where AI helps and where it fails at math

• Where it helps: generating practice problems at a chosen level, scaffolded hints that don't give the answer, alternative explanations of the same concept, and worked examples for students who need another way in.
• Where it fails: multi-step arithmetic, novel or trick problems, and "show your work" that looks right but has a subtle error. Demo one: ask for a multi-step solution, then have the room find the mistake.
• The move: use AI to generate and explain, not to answer. And turn its unreliability into a skill, students verify AI's solutions, which is exactly the error-analysis you already teach.

Watch Keep the demo to one clean wrong answer. Don't turn it into an AI-bashing session; the point is calibrated trust.

Hands-on · build a set and a trap (laptops open)

Each teacher makes two things for a current unit, in the shared doc:

• A tiered problem set — the same concept at three levels (support, on-level, extension), generated with AI and then checked by the teacher. Prompt: "Give me three versions of a two-step equations worksheet, one below grade level, one on level, one extension, with an answer key."
• A "spot the AI's mistake" activity — an AI-produced solution with a planted or real error for students to find and correct, teaching error analysis and healthy skepticism.

Do Teachers verify the AI's answer keys before using them, that verification is the whole lesson. Pairs trade and try to break each other's sets.

Watch Remind them: never hand students an AI answer key you haven't checked.

Share-out + commit

Do A couple of teachers show their tiered set or their trap. Commitment: use one with students this month. Preview: December is the mid-year showcase, bring a win to share.