Reasoning LLMs trigger reflection when their internal uncertainty is high

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

• what is the underlying mechanism of self-reflection in reasoning LLMs?question0.853
  Open question motivating the entire paper; identified as not yet well understood
• The underlying mechanism of self-reflection in reasoning LLMs is not yet well understoodquestion0.853
  Broad gap motivating the entire paper
• A linear reflection direction exists in reasoning LLMs' latent representation space that governs self-reflection behaviorclaim0.803
  Core claim of ReflCtrl that a single direction captures and controls reflection
• Reflection in LLMsconcept0.801
  The core phenomenon studied: the ability of LLMs to evaluate and revise their own reasoning.
• Self-reflection consumes 25-30% of total reasoning tokens in reasoning LLMsclaim0.790
  Empirical observation motivating the need to control reflection for inference efficiency
• LLMs can predict their own responses more accurately than external observers, implying privileged internal knowledgefinding0.790
  Binder et al. finding cited as evidence that LLMs possess introspective capacity analogous to mindfulness
• LLMs linearly represent truth-relevant information beyond the plausibility of text, as evidenced by probes trained on likely performing poorly on anti-correlated datasetsclaim0.785
  Establishes that the observed linear structure is not merely a representation of text probability
• Current LLMs cannot faithfully represent transformative experiences with epistemically opaque outcomes.claim0.778