framework
active
framework:span-representation-analysisSpan Representation Analysis
Framework for characterizing span-level information of sequences of representations, independent of any consciousness estimate; used as a comparison baseline.
Neighborhood — ranked by edge-count
Methods (1)
method
- Method concatenating boundary token vectors, their element-wise product, and difference to form span-level representations from (C)ARR.
Concepts (2)
concept
- The primary paper being extracted — applies IIT 3.0 and 4.0 to LLM representation sequences derived from ToM test data to investigate whether consciousness phenomena can be observed.
- Criterion 3: IIT estimates must achieve higher mean AUC than Span Representation for ToM score classification.associated_withThird of three operational criteria; distinguishes consciousness from inherent LLM representational separations.
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.
- Extracting embeddings from instruction and example spans.
- The central question of whether representational geometry implies corresponding computational structure
- Obtain instruction and example span embeddings at layer L* with chosen pooling.
- A correlational similarity method compared against MAS; uses RDM correlations between model representations.
- How a neural network encodes a semantic concept internally, argued to be better captured by manifolds than by atomic features.
- The idea that features are encoded as directions in activation space.
- Substrate on which causal emergence was computed across agent lifetimes; aligned with learning success.
- Used to visualize LLM true/false representations, revealing clear linear structure separating true from false statements