Model's uncertainty information is encoded in the reflection direction

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

• We hypothesize that explicitly instructing the model to evaluate the correctness of the given statement may change the geometry of truth directions.hypothesis0.776
  Motivating hypothesis for Section 5's investigation of prompt template effects.
• Where and how is information stored in model-internal representations?question0.769
  Core question motivating interchange intervention and interpretability research supported by pyvene
• What is the effect of model instructions on truth directions?question0.759
  Research question motivating Section 5.
• What if the concept being manipulated does not lie on a straight line in the model's representations?question0.756
  The motivating question that opens the paper and leads to the development of manifold steering.
• Reflection-inducing directions emerge more clearly in higher layers (ℓ>5) for both models and datasetsfinding0.752
  Empirical observation about which network layers encode reflection-relevant information.
• Does the model internally maintain a form of 'consistency score' or probability mass over coherent reasoning trajectories, and how is this score modulated during reflection?question0.752
  Promising future research direction about the internal mechanism of error detection.
• The model tends to reflect more when the question is difficult, and accuracy is generally lower for harder questionshypothesis0.750
  Hypothesis explaining negative correlation between reflection rate and accuracy without implying reflection is harmful
• Does instructing the model to assess correctness affect the geometry of truth directions?question0.749
  One of the three guiding research questions of the paper.