concept
active
concept:causal-abstraction-over-cyclic-tasksCausal Abstraction over Cyclic Tasks
Neighborhood — ranked by edge-count
Communities (1)
community
- Neural Geometrymembers_of
Concepts (2)
concept
- Causal abstractionrelated_toA framework the paper uses alongside feature geometry to deepen mechanistic understanding of LMs
- Base-10 Addition Mechanism in Llamaassociated_with
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.
- Language model reasoning tasks with cyclic geometric structure used to test manifold steering.
- Formal definition: H is a constructive abstraction of L under alignment Π when interchange interventions have equivalent effects at both levels.
- The formal method used to establish that the identified circuit causally mediates the model's cyclic reasoning behavior
- Graded notion of causal abstraction measured by IIA; when IIA is alpha < 100%, the model is alpha-on-average approximately abstract.
- Adaptation of Hewitt and Liang control tasks to CausalGym: next-token labels replaced with arbitrary tokens to measure method expressivity
- Causal emergence identification tasks can be understood as causal representation learning tasks.claim0.776Authors propose a conceptual mapping between CE identification and CRL.
- Representational dynamics of causal emergence align with reward improvement in most tasks.finding0.769The trajectory of causal emergence through training mirrors the reward improvement curve across the majority of tested environments.
- Language model experimental setting used to test manifold steering.