Keyword-based reflection step identification

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

• The identification of reasoning steps relies on keyword search, which may be model-specific since different models could prefer different reflection cuesclaim0.791
  Limitation acknowledged regarding generalizability of the reflection identification method
• Reflection Enhancement via Activation Additionmethod0.738
  Adding steering vector in forward direction to push model activations toward stronger reflective behavior.
• No Reflectionconcept0.734
  Reflection level where the model is forced to output an answer immediately without revisiting reasoning.
• Self-reflectionconcept0.726
  The ability of reasoning LLMs to review and revise previous reasoning steps during inference
• Triggered Reflectionconcept0.722
  Reflection level where explicit cue words (e.g., 'wait') prompt the model to inspect and revise reasoning.
• Situational Reflectionconcept0.719
  The specific form of reflection studied, where a model reflects on reasoning generated by another source.
• Reflection is not merely a behavioral artifact of prompting but a phenomenon encoded in the model's activation space.claim0.716
  Central interpretive claim of the paper, supported by steering vector experiments.
• How can we systematically identify effective reflection trigger instructions, rather than relying on trial-and-error?question0.714
  First key research question motivating the methodology.