Accuracy does not vary linearly with latent reflection directions; instead it follows a more non-linear mapping that requires deeper theoretical treatment.

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

• Whether conclusions about latent reflection directions generalize to larger LLMs, different architectures, or broader datasets remains to be verified.question0.846
  Key limitation and open question about experimental scope.
• A linear reflection direction exists in reasoning LLMs' latent representation space that governs self-reflection behaviorclaim0.820
  Core claim of ReflCtrl that a single direction captures and controls reflection
• We hypothesize that explicitly instructing the model to evaluate the correctness of the given statement may change the geometry of truth directions.hypothesis0.807
  Motivating hypothesis for Section 5's investigation of prompt template effects.
• Linear truth directions in LLMs are reliable primarily in factual recall cases and break down when truth assessment depends on computing and storing intermediate results.claim0.806
  Central empirical conclusion of the paper about the fundamental limits of truth directions.
• Steering vectors capture latent dimensions of reflective behavior more faithfully than surface-level embedding similarity.claim0.806
  Supported by the instruction discovery experiments comparing steering vs. embedding baselines.
• Does instructing the model to assess correctness affect the geometry of truth directions?question0.804
  One of the three guiding research questions of the paper.
• Reflective reasoning requires late-stage integration of semantic and reasoning signals, hence reflection-related directions emerge more clearly in higher network layers.claim0.801
  Interpretive claim about the locus of reflection in transformer architecture.
• Reflection-inducing directions emerge more clearly in higher layers (ℓ>5) for both models and datasetsfinding0.798
  Empirical observation about which network layers encode reflection-relevant information.