claim
active
claim:predictors-must-infer-causality-from-correlational-cues-because-tools-never-state-the-relationship-between-feature-and-behaviorPredictors must infer causality from correlational cues because tools never state the relationship between feature and behavior
Mechanism for null result.
Source paper
extracted_from(2026) · Adam Karvonen · Euan Ong · Subhash Kantamneni · Samuel Marks
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (3)
finding
- No activation tool (AO, NLA, SAE) improves on transcript-only baseline for Gemma-3-27B-IT targetsupportsMain negative result; baseline AUROC ~81 with tools within ~1 point.
- Explicit 'X causes Y' statement appears in only 0.6% (3 of 479) of positive claims' NLA outputssupportsTools rarely state causal relationships.
- Soft clues redundant with transcript.
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.
- How can we reliably distinguish causal from merely correlated features when probing neural networks?question0.779Probing open question.
- Mere correlation of feature directions, as used in prior work, need not suffice to find manifoldsclaim0.773Contrast with Engels et al.
- Probe limitation.
- Stated limitation.
- Necessary condition for connectionist cognition.
- Opening question: does the rich geometric structure of neural representations have a causal role in behavior?
- DAS finds causal effect at all training timesteps including when model is just initialisedfinding0.755Corroborates Wu et al. 2023 finding that DAS expressivity inflates causal effect estimates