Probe the signal, test the shortcut · teacher notes

Year 12 · 55 minutes · Live trained probe on synthetic data

Train a logistic probe, flip a shortcut only in held-out data, and compare sequence mean with covariance features.

Student paper journal · Interactive notebook · Editable Python

Preparation and access

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. Keep an individual record even when you investigate in pairs. Record the exact settings, source and data split.

Teaching sequence

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. Train/test split is fixed before fitting. Threshold tuning on held-out data consumes that set for development; a final claim would require a new untouched test set.

Supplied starting example — reveal after prediction

Settings: {'features': 0, 'shift': 1, 'threshold': 0.5, 'seed': 32}

The probe is trained on the first 160 seeded cases only. A shortcut flips on the 80 held-out cases when distribution shift is enabled.

split n TP FN FP TN accuracy brier
Training 160 78 0 0 82 1.0 0.00531636548185332
Held out 80 0 39 41 0 0.0 0.7244117520951743

The table shows up to eight rows. Inspect the notebook for all values, controls and denominators.

Method and scope: Logistic probe trained by gradient descent with L2 penalty. Feature modes: signal+shortcut; sequence means (both zero); centred cross-covariance+constant. These are synthetic representations, not measured LLM activations.

Assessment and feedback

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: Add a validation split, tune once, then evaluate on a separately seeded test set.

Research connections and separate classroom tasks

Covariance-based Sequence Pooling

Read the source

System and data: NTv3 gene ontology and genomic-track tasks

Method: Compares mean pooling with compressed second-order sequence statistics using downstream probes; regularisation matters when labels are scarce.

Task: Compare mean and centred covariance features on sequences with matched means. Record the source model/task, a baseline, the changed factor, a measurement and an alternative explanation. State exactly which part your notebook investigates and which part it does not reproduce.

Boundary: The source method uses second moments and approximations; our centred-covariance experiment is a teaching analogue. It does not recover sequence order.

Explaining 4.2 million genetic variants with state-of-the-art, interpretable predictions

Read the source

System and data: Evo 2 embeddings and labelled genetic variants

Method: Fits variant-effect predictors and constructs evidence-based mechanistic hypotheses; abstract and research report were available for review.

Task: Ask which independent labels and experiments validate a representation-based prediction. Record the source model/task, a baseline, the changed factor, a measurement and an alternative explanation. State exactly which part your notebook investigates and which part it does not reproduce.

Boundary: Genomic predictions are hypotheses, not clinical conclusions. The linked preprint abstract and Goodfire report were reviewed; full preprint text was unavailable.

Using Interpretability to Identify a Novel Class of Alzheimer's Biomarkers

Read the source

System and data: Pleiades and cell-free DNA pilot cohorts

Method: Uses model interpretation to propose fragment-length signals, then tests a simpler classifier on an independent cohort with uncertainty intervals.

Task: Distinguish exploratory feature discovery from independent cohort evaluation and uncertainty. Record the source model/task, a baseline, the changed factor, a measurement and an alternative explanation. State exactly which part your notebook investigates and which part it does not reproduce.

Boundary: This is a small-cohort biomedical study, not a diagnostic classroom tool or an LLM result.

Deploying Interpretability to Production with Rakuten: SAE Probes for PII Detection

Read the source

System and data: Llama 3.1 8B sidecar, English/Japanese synthetic training and real production tests

Method: Compares activation, attention and frozen-SAE probes for token-level PII detection across synthetic-to-real shift, label noise and language settings.

Task: Compare training and shifted test distributions, thresholds and error types. Record the source model/task, a baseline, the changed factor, a measurement and an alternative explanation. State exactly which part your notebook investigates and which part it does not reproduce.

Boundary: Results depend on data and baselines; SAEs are not always superior. Classroom fixtures contain no real personal information.

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.