Attention probes for belief decoding

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

• Attention probing can serve as an efficient tool for detecting performative reasoning and enabling adaptive computation in reasoning modelshypothesis0.821
  Forward-looking hypothesis positioned as a conclusion and future direction of the paper
• attention computationconcept0.777
  Process using Q, K, V to compute a heat map over K and weighted sum of V.
• thought detectionconcept0.761
  Task of detecting a model's internal thoughts; found by Lindsey (2026) to peak at ~2/3 depth in transformers.
• Linear Probe for Evaluation Awarenessmethod0.760
  Nguyen et al. trained linear probes on activations to distinguish evaluation from deployment scenarios.
• How can mechanistic interpretability methods automatically identify attention computations that span multiple attention heads?question0.757
  Long-standing bottleneck in mechanistic interpretability that VPD addresses by working natively on attention weight matrices.
• Sampled-decoding self-reportmethod0.754
  Temperature=0.8 sampled decoding for self-report; reduces collapse moderately but remains discrete and noisy
• Attention Decayconcept0.754
  Decrease in attention paid to system prompt over conversational turns, leading to persona fidelity degradation (cited from Li et al.)
• False Belief Taskmethod0.751
  Classic ToM test requiring understanding that another agent holds a belief different from reality; scored 0/1.