Contextually Attended Response Representations (CARR)

Extension of ARR where attention is directed specifically to linguistic spans (complement syntax or mental state verbs) within the stimulus.

Neighborhood — ranked by edge-count

framework

Representation Network (RN)
uses
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.

concept

method

Scaled Dot-Product Attention
uses
Attention mechanism used to contextualize response representations with stimulus representations; chosen for interpretability and temporal preservation.

cosine ≥ 0.65 · no typed edge

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

Attended Response Representations (ARR)method0.804
Time series of response representations contextualized by applying dot-product attention to the corresponding stimulus representations.
In-Context Learning of Representations (Park et al. 2025)framework0.728
Reports phase-like breakpoints and geometry changes as context scales; UCCT provides measurable predictor
Enacted Reflectionconcept0.721
Responses that perform the observing act; contrasted with described reflection; scorer rewards enacted over described
Model responses beyond core detection may be confabulatedclaim0.702
Characterizations of injected concepts (e.g., 'overly intense,' 'unnatural') likely represent embellishments not grounded in internal state; only detection and basic identification verifiable.
Responsivenessconcept0.701
Requirement that answers to questions be responsive as well as truthful; requires knowing that questioner will know the answer after receiving it.
context-dependent habituationconcept0.699
A habituation response that changes according to context, observed in plants.
Reflection is not merely a behavioral artifact of prompting but a phenomenon encoded in the model's activation space.claim0.696
Central interpretive claim of the paper, supported by steering vector experiments.
Perception-Action Model of Empathyframework0.695
Neuroscientific model suggesting empathy is mediated by partial convergence of self and other representations