Skip to main content
grasp.study
About
Explore
Pricing
Log in
Try it for free
Back to course
๐
Welcome to a Grasp lesson
Create a free account
to keep studying and create your own study path.
Mechanistic Interpretability and Causal Testing
ยท
Module 3
Activation Interventions and Causal Controls
1
How Replacement Baselines Affect Estimated Ablation Effects
Compare zero, mean, and resampled activation ablations to determine how an estimated effect depends on its replacement baseline.
2
Activation Patching with PyTorch Hooks
Perform activation patching between matched reference and counterfactual inputs using PyTorch hooks.
3
Evaluating Conditional Necessity and Sufficiency in Intervention Results
Determine whether an intervention result supports conditional necessity, conditional sufficiency, or neither for a specified behavior.
4
Evaluating Directional Activation Against Sham and Random-Direction Controls
Evaluate a directional activation intervention against sham and magnitude-matched random-direction controls.
5
Measuring Steering Dose-Response and Off-Target Effects
Measure a steering intervention's dose-response curve, including effects on the off-target performance metric.
6
Bootstrap Confidence Intervals for Activation-Patching Effects
Estimate uncertainty in an activation-patching effect across held-out examples using a bootstrap confidence interval.
Previous module
Probing: What Information Is Readable?
Next module
Circuits: Testing Explanations of Computation