finding
active
finding:within-task-variance-of-model-performance-is-large-circuits-approximate-average-case-but-poorly-individual-data-pointsWithin-task variance of model performance is large: circuits approximate average case but poorly individual data points
Undermines concept-based task definition in circuit discovery.
Source paper
extracted_from(2025) · Lee Sharkey · Bilal I. Chughtai · Joshua Batson · Jack W. Lindsey +25
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Lists issues: concept-based tasks, flawed decomposition, low faithfulness.
Related by similarity (8)
cosine ≥ 0.65 · no typed edgeEntities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.
- Fig. 23.
- When task gradient norms differ greatly, large-norm tasks have not converged while small-norm tasks have nearly convergedhypothesis0.772Motivates setting αk = max norm to enable further learning on under-converged tasks
- Hypothesis box in §3.2.
- Supports the claim against single-layer probing approaches used in prior work.
- Fig. 22.
- Core framing.
- Supported by empirical comparison showing VS achieves KL divergence of 0.12 from pretraining distribution vs. 14.89 for direct prompting
- Data-induced competition over resources.