Research map / Learning theory and algorithms / Prediction and decision losses
Provably Tracking Equivalent Mechanistic Interpretations Across Neural Networks
Alan Sun, Mariya Toneva
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| compute congruity between models with same vs different interpretation on n-permutation detection | Congruity · 10-permutation detection, 6 hard-coded transformers | congruity: within-group 0.43 vs across-group 0.01 | Improved |
| compare Pythia vs GPT2 models on IOI with Congruity | Congruity · IOI, 85M to 2.9B parameters | congruity: across-group 0.13 vs within 0.73 and 0.92 | Improved |
| restrict next-token prediction to terminal punctuation or closing brackets/quotations | GPT2 vs POS circuit · C4 next-token prediction | congruity with POS: all-token control 0.48; medium effect size | Improved |
| restrict next-token prediction to articles or prepositions | GPT2 vs POS circuit · C4 next-token prediction | congruity with POS | No clear effect |
Closest papers by meaning
- Validating Mechanistic Interpretations: An Axiomatic Approach (ICML 2025 · similarity 0.91)
- Everything, Everywhere, All at Once: Is Mechanistic Interpretability Identifiable? (ICLR 2025 · similarity 0.91)
- MIB: A Mechanistic Interpretability Benchmark (ICML 2025 · similarity 0.88)
- Towards Global-level Mechanistic Interpretability: A Perspective of Modular Circuits of Large Language Models (ICML 2025 · similarity 0.87)
- The Non-Linear Representation Dilemma: Is Causal Abstraction Enough for Mechanistic Interpretability? (NeurIPS 2025 · similarity 0.87)
- Extracting Interpretable Task-Specific Circuits from Large Language Models for Faster Inference (AAAI 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
