// ⌖ Algorithms Atlas

Interpretability

Opening the black box: attribution, probing, superposition, sparse autoencoders, circuit tracing and steering.

7 core2 extended2016–2025

TIMELINE

2016 Grad-CAM2017 Feature Attrib2020 Logit Lens & P2022 Circuits & Ind2022 Superposition2022 Locating & Edi2023 Sparse Autoenc2023 Activation Ste2025 Circuit Tracin

01Core algorithms

Learn these first. Sorted by year.

Feature Attribution (Integrated Gradients / SHAP) Measures how much each input feature contributed to a prediction. 2017core Logit Lens & Probing Decodes intermediate layers into tokens, or probes them with small classifiers. 2020core Circuits & Induction Heads Finds circuits implementing algorithms, like induction heads that copy patterns. 2022core Superposition Models pack more features than dimensions by superimposing them. 2022core Sparse Autoencoders (SAE) Decomposes activations into many sparse, interpretable features. 2023core Activation Steering Adds a direction vector to activations at inference to change behavior. 2023core Circuit Tracing & Attribution Graphs Maps causal feature pathways behind a specific response. 2025core

02Extended

Variants and follow-ups worth knowing once the core is clear.

Grad-CAM Gradient-weighted heatmaps showing where a CNN looks. 2016ext Locating & Editing Facts (ROME) Locates where facts are stored and edits those weights directly. 2022ext

03Course units that teach these

U91 Interpretability: from Attribution to Circuits Attribution, probes, the logit lens, superposition, SAEs, circuit tracing, knowledge editing and activation steering. 10 h9 algorithms4 paperslesson ready