Year 6 · 35 minutes · Explicit teaching model
Help Pip weigh fictional weather clues. Move the starting guess, clue reliability and number of clues; the probabilities update live.
Student paper journal · Interactive notebook · Editable Python
Try the starting controls, save two runs and reopen a downloaded project. Print the student journal before class. If Python cannot load, use the supplied example below for a prediction and evidence critique; rerunning changed inputs requires the live or native notebook. Do not present paper discussion as a new model experiment. Draw, point or explain aloud. A partner or adult may record your words.
Start with 5–8 minutes of prediction, use about half the lesson for paired comparisons, then reserve at least 10 minutes for a counterexample and an individual explanation. Use “the teaching model updates a probability”, not “the computer feels worried”. Discuss why copying one cloud photograph three times does not create three independent clues.
Settings: {'prior': 0.5, 'reliability': 0.7, 'clues': 3, 'story': 0}
Pip is planning a picnic. After 3 clues, this teaching model gives rain probability 70.0%. It updates a number; it has no feelings.
| clue | p_rain | evidence |
|---|---|---|
| 0 | 0.5 | Starting guess |
| 1 | 0.7 | Dark clouds support rain |
| 2 | 0.844828 | Dark clouds support rain |
| 3 | 0.7 | Clear sky supports dry weather |
The table shows up to eight rows. Inspect the notebook for all values, controls and denominators.
Method and scope: Updated odds = previous odds × likelihood ratio. Clues are assumed independent given the weather; repeated copies of one clue would violate this assumption.
Assess four criteria, each 0–2: testable prediction; comparison identifying what stayed fixed; accurate use of exact evidence; counterexample and limited conclusion. 0 means absent or contradicted by the record, 1 means partly supported, and 2 means clear and supported. A surprising result earns no penalty.
Beginning response: ‘The picture looks right.’ Ask for a particular case. Developing response: names one number without its setting. Ask what comparison supports it. Secure response: states the setting and measured change and separates the notebook's result from the source paper's claim. Extending response: authors or reserves a new case, tests an alternative explanation, and revises the claim if needed. These are marking examples, not pupil data.
Extension: Reverse the clue order and compare the final probability under the independence assumption.
System and data: Three interpretability lecture outlines
Method: Introduces causal abstraction, circuits and learning in context; a teaching guide rather than one dataset.
Task: Describe a clue, a guess and a test in different sentences. Use the linked notebook to make two observations. Draw or describe one result and one thing this activity cannot tell us about the source system.
Boundary: This entry reviews the article and lecture outlines, not a full transcription of the videos.
System and data: Many-shot context and activation-steering experiments
Method: Fits a Bayesian account that distinguishes likelihood updates from changes to priors; tests how context strength changes steering effects.
Task: Change the starting guess, then reveal the same clues. Use the linked notebook to make two observations. Draw or describe one result and one thing this activity cannot tell us about the source system.
Boundary: A fitted account of model behaviour does not establish that a model has human beliefs.
System and data: Gemma-2-2B activations from Pile-Uncopyrighted
Method: Compares temporal feature analysis with ReLU, TopK and BatchTopK SAEs, testing predictable versus innovation components and event structure.
Task: Reveal story clues one at a time and record each change. Use the linked notebook to make two observations. Draw or describe one result and one thing this activity cannot tell us about the source system.
Boundary: Our Bayesian story model is explicit and hand-specified; it is not Temporal Feature Analysis applied to an LLM.
System and data: DeepSeek-R1 families and GPT-OSS-120B on MMLU-Redux and GPQA-Diamond
Method: Trains context-pooling probes to forecast eventual answers during reasoning and evaluates early-exit token cost versus benchmark accuracy.
Task: An early guess can match a later answer and still be wrong. Use the linked notebook to make two observations. Draw or describe one result and one thing this activity cannot tell us about the source system.
Boundary: Savings depend on task and paper version; the Goodfire post and later arXiv revision report different percentages. Our trajectories are simulated.
System and data: Llama 3.1 8B and SimpleStories
Method: Collects sentence-end activations, fits trajectories and compares emotional organisation with behavioural readouts and human valence/arousal judgements.
Task: Plot how clues about a fictional character change as the story unfolds. Use the linked notebook to make two observations. Draw or describe one result and one thing this activity cannot tell us about the source system.
Boundary: Representing a character's emotions does not mean the model has emotions. Classroom story probabilities are hand-specified.
System and data: DeepSeek R1 671B on custom reasoning and OpenR1-Math data
Method: Trains two SAEs and studies feature activations and steering timing/strength; partial features do not expose a complete reasoning transcript.
Task: Change a clue at different times and observe whether the ending changes. Use the linked notebook to make two observations. Draw or describe one result and one thing this activity cannot tell us about the source system.
Boundary: An SAE reveals partial patterns, not a complete transcript of reasoning. Effects do not generalise automatically.
Independent Brightlab adaptations; no Goodfire endorsement or full reproduction claim. Source methods reviewed 7 September 2026. Curriculum connections are selected planning links; confirm your school syllabus.