mlsanityData to 7 Oct 2026

Diagnosing Correctness Probes under Self-Judgement Confounding

arXiv 2607.16799 ↗First posted 18 Jul 2026cs.CL cs.LGcited by 1

Not on the map. Why?

Abstract (arXiv)

Hidden-state readouts can predict whether language-model outputs are correct, but objective correctness (OC) usually agrees with the model's own self-judgement (SJ), leaving the decoded signal semantically ambiguous. We construct conflict cases in which OC and SJ predict opposite readout orderings. On high-confidence disagreements, conventional correctness-labelled contrasts often rank incorrect/self-endorsed responses above correct/self-rejected responses, following SJ rather than OC. We estimate factorial SJ- and OC-associated directions and evaluate their polarity across mathematical reasoning and factual recall. Across four instruction-tuned models up to 14B parameters, the SJ-associated direction transfers above chance in both cross-domain directions for every model, whereas the OC-associated direction has a below-chance point estimate for the expected OC ordering in every corresponding condition. This transfer asymmetry develops across middle-to-late layers, persists under answer-likelihood, sequence-length, and null-direction controls, and extends to MMLU and binary TruthfulQA without target-domain direction fitting. Across the studied models and diagnostic subsets, the most reliably transferable component preserves SJ-associated polarity. Transferability alone therefore does not establish objective-correctness semantics.

Key facts. Cited by 1 paper in the map (data through 7 Oct 2026). Builds on 26 of its 33 references that are in the map. First posted on arXiv 18 Jul 2026.

Builds on

26 of 33 references.

References · 26

  1. 2024The Llama 3 Herd of Modelscited by 14,192
  2. 2021Training Verifiers to Solve Math Word Problemscited by 9,743
  3. 2021Measuring Mathematical Problem Solving With the MATH Datasetcited by 5,203
  4. 2023Let's Verify Step by Stepcited by 3,598
  5. 2021TruthfulQA: Measuring How Models Mimic Human Falsehoodscited by 2,919
  6. 2022Language Models (Mostly) Know What They Knowcited by 1,783
  7. 2023Inference-Time Intervention: Eliciting Truthful Answers from a Language Modelcited by 1,259
  8. 2023Can LLMs Express Their Uncertainty? An Empirical Evaluation of Confidence Elicitation in LLMscited by 946
  9. 2022Discovering Latent Knowledge in Language Models Without Supervisioncited by 733
  10. 2022Teaching Models to Express Their Uncertainty in Wordscited by 731
  11. 2023The Geometry of Truth: Emergent Linear Structure in Large Language Model Representations of True/False Datasetscited by 673
  12. 2021A Diverse Corpus for Evaluating and Developing English Math Word Problem Solverscited by 314
14 more
  1. 2023Cognitive Dissonance: Why Do Language Model Outputs Disagree with Internal Representations of Truthfulness?cited by 62
  2. 2024Latent Space Chain-of-Embedding Enables Output-free LLM Self-Evaluationcited by 56
  3. 2025Language Models Are Capable of Metacognitive Monitoring and Control of Their Internal Activationscited by 37
  4. 2025Language Models Fail to Introspect About Their Knowledge of Languagecited by 28
  5. 2026How do LLMs Compute Verbal Confidencecited by 23
  6. 2025Probing the Geometry of Truth: Consistency and Generalization of Truth Directions in LLMs Across Logical Transformations and Question Answering Taskscited by 18
  7. 2025The Geometries of Truth Are Orthogonal Across Taskscited by 9
  8. 2025When Truthful Representations Flip Under Deceptive Instructions?cited by 9
  9. 2025Enhancing Uncertainty Estimation in LLMs with Expectation of Aggregated Internal Beliefcited by 9
  10. 2025LLM Knowledge is Brittle: Truthfulness Representations Rely on Superficial Resemblancecited by 9
  11. 2026Can LLMs Introspect? A Reality Checkcited by 8
  12. 2025Do LLMs Really Know What They Don't Know? Internal States Mainly Reflect Knowledge Recall Rather Than Truthfulnesscited by 7
  13. 2026Are LLM Uncertainty and Correctness Encoded by the Same Features? A Functional Dissociation via Sparse Autoencoderscited by 5
  14. 2026How LLMs Detect and Correct Their Own Errors: The Role of Internal Confidence Signalscited by 3

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 1.

1 paper.

  1. 2026The Truth Was Never Gone: Perfect Aliasing in Compliant-Context Truth Probescited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025Language Models can perform Single-Utterance Self-Correction of Perturbed Reasoning
  2. 2026When Decodability Is Not Enough: Logical Validity Representations, Behavioral Dissociation, and Causal Tests in Language Models
  3. 2026Confidence Geometry Reveals Trace-Level Correctness in Large Language Model Reasoning
  4. 2026Wrong Before Right: Late Rescue and Interface Failure in Aligned Language Models
  5. 2026Does Inference Scaling Improve Reasoning Faithfulness? A Multi-Model Analysis of Self-Consistency Tradeoffs
  6. 2025Catch Me If You Can: How Smaller Reasoning Models Pretend to Reason with Mathematical Fidelity
  7. 2026Masked by Consensus: Disentangling Privileged Knowledge in LLM Correctness
  8. 2026Locating Answer-Correctness Signals in Frozen Large Language Models