A probe may achieve high performance even on representations that are not causally relevant for the task

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

• A probe achieving high classification accuracy provides no guarantee that the model actually distinguishes those classes in downstream computationsclaim0.861
  Motivation for causal evaluation over purely behavioural probing accuracy
• Even validated probes may capture distributed representations mixing emotive states with correlated features like persona or styleclaim0.816
  Caveat on probe interpretation; does not negate the introspection result but affects interpretation of the target variable
• When probe and self-report agree and move together causally, confidence in both increases as evidence they track the same underlying stateclaim0.814
  Convergent validity logic applied to LLM interpretability; probes validate self-reports and vice versa
• If steering in a purported concept direction does not shift self-report in the expected direction, probe quality becomes suspect, especially when conventional probe metrics alone looked acceptable.quote0.802
  Key methodological insight: introspection enables a new probe validation criterion beyond conventional separation metrics
• Direct probes over learned activations in standard basis may fail to reveal the actual causal role of representations because they are highly distributedclaim0.801
  Supported by the finding that non-trivial rotations are required to find aligned representations.
• Attention probing can serve as an efficient tool for detecting performative reasoning and enabling adaptive computation in reasoning modelshypothesis0.793
  Forward-looking hypothesis positioned as a conclusion and future direction of the paper
• Probes trained under different explicit instruction variants are highly aligned with each other despite different wording.claim0.793
  Shows the key divide is passive vs. active framing, not the specific wording of instructions.
• Simple difference-in-mean probes generalize as well as other probing techniques while identifying directions which are more causally implicated in model outputsclaim0.792
  Key methodological claim: MM probes are both competitive in accuracy and superior in causal influence