method
active
method:attended-response-representations-arrAttended Response Representations (ARR)
Time series of response representations contextualized by applying dot-product attention to the corresponding stimulus representations.
Neighborhood — ranked by edge-count
Frameworks (1)
framework
- Novel construct introduced by this paper: a hypothetical graph embedded in the time series of LLM representations, where each dimension is a node and latent connections are edges.
Concepts (1)
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.
Methods (1)
method
- Attention mechanism used to contextualize response representations with stimulus representations; chosen for interpretability and temporal preservation.
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.
- Extension of ARR where attention is directed specifically to linguistic spans (complement syntax or mental state verbs) within the stimulus.
- Tuples are referenced by matching typed fields, akin to a relational database.
- Representations of one's own mental states; associated with consciousness in higher-order theories.
- Strategy using GPT-4o, Claude 3.5 Sonnet, and Gemini to generate additional responses preserving original meaning, targeting ≥1000 words concatenated per score category.
- Reports phase-like breakpoints and geometry changes as context scales; UCCT provides measurable predictor
- Responses that perform the observing act; contrasted with described reflection; scorer rewards enacted over described
- A response containing multiple distinct attempts to answer the prompt, used as primary metric for ESR
- Justifies the methodological choice of attention over concatenation, mean pooling, residual connections, or joint embedding.