A linear reflection direction exists in reasoning LLMs' latent representation space that governs self-reflection behavior

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

• The underlying mechanism of self-reflection in reasoning LLMs is not yet well understoodquestion0.845
  Broad gap motivating the entire paper
• what is the underlying mechanism of self-reflection in reasoning LLMs?question0.839
  Open question motivating the entire paper; identified as not yet well understood
• Whether conclusions about latent reflection directions generalize to larger LLMs, different architectures, or broader datasets remains to be verified.question0.832
  Key limitation and open question about experimental scope.
• Reflective reasoning requires late-stage integration of semantic and reasoning signals, hence reflection-related directions emerge more clearly in higher network layers.claim0.820
  Interpretive claim about the locus of reflection in transformer architecture.
• Accuracy does not vary linearly with latent reflection directions; instead it follows a more non-linear mapping that requires deeper theoretical treatment.claim0.820
  Theoretical limitation identified by the authors distinguishing reflection from stylistic tasks.
• Reflection is not merely a behavioral artifact of prompting but a phenomenon encoded in the model's activation space.claim0.811
  Central interpretive claim of the paper, supported by steering vector experiments.
• Truthful behavior in LLMs is not confined to a single linear axis; multiple orthogonal directions can independently mediate itclaim0.809
  Central interpretive claim of the paper
• What are the specific attention heads or MLP neurons (circuits) responsible for self-reflection in LLMs?question0.803
  Future research question about pinpointing fine-grained mechanistic components of reflection.