Skip to main content
grasp.study
About
Explore
Pricing
Log in
Try it for free
Back to course
đź‘‹
Welcome to a Grasp lesson
Create a free account
to keep studying and create your own study path.
Mechanistic Interpretability and Causal Testing
·
Module 1
From Internal Concepts to Testable Claims
1
Tracing Attention and MLP Contributions Through a Transformer Block
Trace the residual-stream contribution of attention and MLP components through a supplied transformer block.
2
Units of Explanation in Neural Networks
Distinguish neurons, activation directions, sparse-autoencoder latents, and circuits as different proposed units of explanation.
3
From Internal Representation to Testable Prediction
Translate a claim that an internal feature represents a concept into a falsifiable prediction.
4
Constructing Matched Examples to Isolate Shortcut Cues
Construct matched examples that vary a candidate shortcut cue while preserving task-relevant evidence.
5
Selecting Target and Off-Target Metrics for Interpretability Experiments
Choose a target-behavior metric and an off-target performance metric for an interpretability experiment.
Previous module
Next module
Probing: What Information Is Readable?