Reflection is not merely a behavioral artifact of prompting but a phenomenon encoded in the model's activation space.

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

• A linear reflection direction exists in reasoning LLMs' latent representation space that governs self-reflection behaviorclaim0.811
  Core claim of ReflCtrl that a single direction captures and controls reflection
• Enacted reflection may correspond to silent mid-layer processing; described reflection to the motor impulse of concepts leaking through to output.claim0.810
  Mechanistic analog connecting Lindsey's layer-localized findings to the scorer's enacted/described distinction
• Reflective reasoning requires late-stage integration of semantic and reasoning signals, hence reflection-related directions emerge more clearly in higher network layers.claim0.792
  Interpretive claim about the locus of reflection in transformer architecture.
• When does the model initiate reflection during its reasoning process?question0.792
  First central research question motivating ReflCtrl investigation
• Activation steering of reflection has dual-use implications: it can enhance reflection as a defense mechanism, but malicious actors may also use it to inhibit reflection to facilitate jailbreaks.claim0.787
  Applied dual-use conclusion drawn from the paper's findings.
• Contrasting No Reflection with Triggered Reflection (µ(0→2)) provides a stronger reflection signal than contrasting Intrinsic with Triggered Reflection (µ(1→2)).claim0.784
  Empirical interpretation of which reference baseline yields more useful steering vectors.
• Reflection does not only emerge in SFT or RL stages but arises earlier during pre-training.claim0.783
  Cited finding from Shah et al. contextualizing the training origins of reflection.
• What real phenomenon is reflected in these experiments?question0.781
  Asks what underlying reality causes the consistent choices.