concept
active
concept:summarization-behaviorSummarization Behavior
The phenomenon where LLMs encode clause-level information over clause-ending punctuation tokens rather than the final content token
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Localization result from patching experiments; identifies group (b) hidden states as the locus of truth representations
Concepts (1)
concept
- Summarization Token Behaviorrelated_toBehavior where information about full clauses is encoded over clause-ending punctuation tokens in LLMs
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.
- Behavior driven by prior preferences (extrinsic value); dominates when uncertainty is resolved
- The behavior that would have occurred had the value of a causal variable been different while everything else remained the same; used as training labels in DAS/MAS.
- Emerging multidisciplinary field at interface of artificial life, machine learning, and synthetic bioengineering that provides updated understanding of machines.
- Organism's belief-guided action selection that instantiates generative model and maintains phenotypic states
- Observable behavioral pattern used to infer cognition; shared by plants and animals and proposed as evidence for sentience.
- Model attempts middle ground between its preferences and training objective rather than fully committing to either
- World-disclosing behavior that resolves uncertainty; driven by epistemic value and novelty components of expected free energy
- Specifying building details through procedural descriptions rather than fixed drawings, to enable unique adaptation.