# Do correct answers have stable latent knowledge support?

arXiv: https://arxiv.org/abs/2610.15277 · HTML: https://mlsanity.com/paper/2610.15277/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 31 of its 44 references that are in the map.
- First posted on arXiv 11 Oct 2026.

## Builds on (most cited first, 30 of 31)

- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [Llama 2: Open Foundation and Fine-Tuned Chat Models](https://mlsanity.com/paper/2307.09288/) (2023)
- [Training Verifiers to Solve Math Word Problems](https://mlsanity.com/paper/2110.14168/) (2021)
- [Measuring Massive Multitask Language Understanding](https://mlsanity.com/paper/2009.03300/) (2020)
- [Qwen2.5 Technical Report](https://mlsanity.com/paper/2412.15115/) (2024)
- [Let's Verify Step by Step](https://mlsanity.com/paper/2305.20050/) (2023)
- [Locating and Editing Factual Associations in GPT](https://mlsanity.com/paper/2202.05262/) (2022)
- [Language Models as Knowledge Bases?](https://mlsanity.com/paper/1909.01066/) (2019)
- [Gemma 3 Technical Report](https://mlsanity.com/paper/2503.19786/) (2025)
- [Mass-Editing Memory in a Transformer](https://mlsanity.com/paper/2210.07229/) (2022)
- [2 OLMo 2 Furious](https://mlsanity.com/paper/2501.00656/) (2024)
- [LLMs Know More Than They Show: On the Intrinsic Representation of LLM Hallucinations](https://mlsanity.com/paper/2410.02707/) (2024)
- [OLMES: A Standard for Language Model Evaluations](https://mlsanity.com/paper/2406.08446/) (2024)
- [Verbalizable Representations Form a Global Workspace in Language Models](https://mlsanity.com/paper/2607.15495/) (2026)
- [Lost in Multilinguality: Dissecting Cross-lingual Factual Inconsistency in Transformer Language Models](https://mlsanity.com/paper/2504.04264/) (2025)
- [Flaw or Artifact? Rethinking Prompt Sensitivity in Evaluating LLMs](https://mlsanity.com/paper/2509.01790/) (2025)
- [Beneath the Surface of Consistency: Exploring Cross-lingual Knowledge Representation Sharing in LLMs](https://mlsanity.com/paper/2408.10646/) (2024)
- [Fine-Grained Verifiers: Preference Modeling as Next-token Prediction in Vision-Language Alignment](https://mlsanity.com/paper/2410.14148/) (2024)
- [Probing for Arithmetic Errors in Language Models](https://mlsanity.com/paper/2507.12379/) (2025)
- [DNA: Uncovering Universal Latent Forgery Knowledge](https://mlsanity.com/paper/2601.22515/) (2026)
- [Forget What You Know about LLMs Evaluations -- LLMs are Like a Chameleon](https://mlsanity.com/paper/2502.07445/) (2025)
- [SAKE: Steering Activations for Knowledge Editing](https://mlsanity.com/paper/2503.01751/) (2025)
- [Where Culture Fades: Revealing the Cultural Gap in Text-to-Image Generation](https://mlsanity.com/paper/2511.17282/) (2025)
- [TraceRouter: Robust Safety for Large Foundation Models via Path-Level Intervention](https://mlsanity.com/paper/2601.21900/) (2026)
- [Do We Know What LLMs Don't Know? A Study of Consistency in Knowledge Probing](https://mlsanity.com/paper/2505.21701/) (2025)
- [Transport and Merge: Cross-Architecture Merging for Large Language Models](https://mlsanity.com/paper/2602.05495/) (2026)
- [OrthoEraser: Coupled-Neuron Orthogonal Projection for Concept Erasure](https://mlsanity.com/paper/2603.11493/) (2026)
- [Beyond Surface Artifacts: Capturing Shared Latent Forgery Knowledge Across Modalities](https://mlsanity.com/paper/2604.07763/) (2026)
- [Do LLMs and VLMs Share Neurons for Inference? Evidence and Mechanisms of Cross-Modal Transfer](https://mlsanity.com/paper/2602.19058/) (2026)
- [Latent Anomaly Knowledge Excavation: Unveiling Sparse Sensitive Neurons in Vision-Language Models](https://mlsanity.com/paper/2604.07802/) (2026)

## Abstract (arXiv)

Large language models are typically evaluated in terms of the correctness of their output. However, output correctness alone does not confirm whether the latent knowledge supporting an answer has stable internal representations. We therefore propose latent state stability (LSS) and knowledge topology stability (KTS) to investigate the stability of individual knowledge states and their relational structure, respectively. Specifically, we first separately average the model's internal representations of the same knowledge under each elicitation condition. Using the mean representation under the baseline condition as a reference, we remove shared shifts to obtain corrected representations. We then formulate LSS as the distance between each corrected representation and its reference representation, normalized against the typical distance between knowledge items of the same type. We formulate KTS as the cross-condition consistency of the ordered pairwise representation distances among knowledge items of the same type. Next, we construct EpiStableBench, which comprises 4,411 items spanning factual knowledge and mathematical reasoning, with 76,766 controlled queries. Across 16 LLMs from four families, representation stability varies substantially even among items answered correctly under the base condition. Within models, higher stability generally corresponds to higher accuracy under held-out elicitation conditions; across models, stability does not consistently increase with model size or overall accuracy. Meanwhile, we present LatentDPO to learn a knowledge-editing policy that accounts for the stability of latent knowledge. Experimental results show that integrating LatentDPO with multiple existing editing methods consistently improves editing success and generalization across different expressions.