On a synthetic Lean tactic-prediction task, exact circuit edge lists barely overlap across dense and sparse checkpoints while attention-head sets and size rankings do, so a circuit claim is well defined only once graph, threshold, and comparison level are stated.
Sparse interventions in language models with differentiable masking
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Circuit Claims Depend on What Is Extracted and How It Is Compared
On a synthetic Lean tactic-prediction task, exact circuit edge lists barely overlap across dense and sparse checkpoints while attention-head sets and size rankings do, so a circuit claim is well defined only once graph, threshold, and comparison level are stated.