LLumen
50 minutes

Teaching the Lumen Labs

Facilitate decision trees, classifiers, bias, hallucination, prompt, privacy, and tradeoff labs.

Objectives

  • Run a 40-minute lab period with clear timing
  • Debrief accuracy vs fairness without shaming students
  • Connect lab outcomes to exit tickets and mastery

Facilitation pattern

Predict → do → compare → explain. Always ask what the model optimized and who might be harmed when it is wrong.

Common stuck points

Students equate high accuracy with ‘smart’ and fair. Use the rare-event fire drill example and disaggregated metrics. On generative labs, students over-trust fluent prose — force claim-by-claim verification.

Assessment

Score process: hypothesis quality, metric interpretation, model card honesty, and verification plans — not only final accuracy or pretty prompts.