mlsanityData to 14 Oct 2026

Hallucination detection from model internals

23 papers1 this weekHigh confidence in this grouping: the papers form one clear topicHigh confidence in this grouping: the papers form one clear topic · About

Keywords: hallucination, hallucinations, steering, detection, lvlms

Key facts. Hallucination detection from model internals: 23 papers on arXiv since 2023 (data through 14 Oct 2026). 4 of its 23 mapped papers (17%) cite “Do I Know This Entity? Knowledge Awareness and Hallucinations in Language Models” (2024, arXiv 2411.14257). 2 new papers 24 Sep–14 Oct.

What’s new

24 Sep – 14 Oct 2026

2 new since 24 Sep

  1. TRACESTEER: Trace Driven Activation Steering for Hallucination Mitigation in Diffusion Language Models
  2. Where Hallucinations Live: A Cross-Architecture Circuit in VQ-Tokenized Vision-Language Models

Foundations

Most cited by papers here

  1. Do I Know This Entity? Knowledge Awareness and Hallucinations in Language Models2024cited by 4 here
  2. Interpreting and Editing Vision-Language Representations to Mitigate Hallucinations2024cited by 3 here
  3. SAFE: A Sparse Autoencoder-Based Framework for Robust Query Enrichment and Hallucination Mitigation in LLMs2025cited by 2 here
  4. Do Androids Know They're Only Dreaming of Electric Sheep?2023cited by 1 here
  5. ReDeEP: Detecting Hallucination in Retrieval-Augmented Generation via Mechanistic Interpretability2024cited by 1 here

Lineage

The papers this direction builds on, oldest first.

Each paper on the line cites the one before.

  1. 2022Interpretability in the Wild: a Circuit for Indirect Object Identification in GPT-2 smallcited by 3 here
  2. 2023Towards Automated Circuit Discovery for Mechanistic Interpretabilitycited by 2 here
  3. 2023Sparse Autoencoders Find Highly Interpretable Features in Language Modelscited by 7 here
  4. 2024Scaling and evaluating sparse autoencoderscited by 3 here
  5. 2024Jumping Ahead: Improving Reconstruction Fidelity with JumpReLU Sparse Autoencoderscited by 2 here
  6. 2024Gemma Scope: Open Sparse Autoencoders Everywhere All At Once on Gemma 2cited by 5 here
  7. 2024Do I Know This Entity? Knowledge Awareness and Hallucinations in Language Modelscited by 4 here
  8. 2025A Survey on Sparse Autoencoders: Interpreting the Internal Mechanisms of Large Language Modelscited by 2 here
16 more it builds on, not in this chain
  1. 2022Extracting Latent Steering Vectors from Pretrained Language Modelscited by 1 here
  2. 2022Toy Models of Superpositioncited by 2 here
  3. 2023Eliciting Latent Predictions from Transformers with the Tuned Lenscited by 1 here
  4. 2023Steering Language Models With Activation Engineeringcited by 2 here
  5. 2023Towards Vision-Language Mechanistic Interpretability: A Causal Tracing Tool for BLIPcited by 1 here
  6. 2023Representation Engineering: A Top-Down Approach to AI Transparencycited by 7 here
  7. 2023Steering Llama 2 via Contrastive Activation Additioncited by 3 here
  8. 2024Interpreting and Editing Vision-Language Representations to Mitigate Hallucinationscited by 3 here
  9. 2024Semantics-Adaptive Activation Intervention for LLMs via Dynamic Steering Vectorscited by 2 here
  10. 2024Large Multi-modal Models Can Interpret Features in Large Multi-modal Modelscited by 2 here
  11. 2024Sparse autoencoders reveal selective remapping of visual concepts during adaptationcited by 2 here
  12. 2025SAE-V: Interpreting Multimodal Models for Enhanced Alignmentcited by 2 here
  13. 2025SAFE: A Sparse Autoencoder-Based Framework for Robust Query Enrichment and Hallucination Mitigation in LLMscited by 2 here
  14. 2025Sparse Autoencoders Learn Monosemantic Features in Vision-Language Modelscited by 2 here
  15. 2025Learning to Steer: Input-dependent Steering for Multimodal LLMscited by 2 here
  16. 2025SteerVLM: Robust Model Control through Lightweight Activation Steering for Vision Language Modelscited by 2 here

The 24 most cited of 56.

Browse

18 of 23, ranked 6–23

  1. 6Dynamic Multimodal Activation Steering for Hallucination Mitigation in Large Vision-Language Models20261
  2. 7Steering LVLMs via Sparse Autoencoder for Hallucination Mitigation20251
  3. 8SAVE: Sparse Autoencoder-Driven Visual Information Enhancement for Mitigating Object Hallucination20251
  4. 9Toward Faithful Retrieval-Augmented Generation with Sparse Autoencoders20251
  5. 10AFTER: Mitigating the Object Hallucination of LVLM via Adaptive Factual-Guided Activation Editing20260
  6. 11One-shot Optimized Steering Vector for Hallucination Mitigation for VLMs20260
  7. 12Towards Interpretable Hallucination Analysis and Mitigation in LVLMs via Contrastive Neuron Steering20260
  8. 13Detection Without Correction: A Robust Asymmetry in Activation-Based Hallucination Probing20260
  9. 14HalluSAE: Detecting Hallucinations in Large Language Models via Sparse Auto-Encoders20260
  10. 15Dual-Pathway Circuits of Object Hallucination in Vision-Language Models20260
All 18
  1. 16Hallucination Is Linearly Decodable from Mid-Layer Hidden States in Quantized LLMs20260
  2. 17Steer Where It Matters: Token-Level Visual-Sensitivity Steering for LVLMs Hallucination Mitigation20260
  3. 18Whisper Hallucination Detection and Mitigation via Hidden Representation Steering and Sparse AutoEncoders20260
  4. 19Prompt Embedding Probes (PEP): Hallucination Detection in LLMs from Hidden States20260
  5. 20Gated Activation Steering for Reducing Sycophancy & Hallucination in Medical Question Answering20260
  6. 21Hallucination reduction with CASAL: Contrastive Activation Steering For Amortized Learning20250
  7. 22Adaptive Activation Steering: A Tuning-Free LLM Truthfulness Improvement Method for Diverse Hallucinations Categories20240
  8. 23Beyond Logit Lens: Contextual Embeddings for Robust Hallucination Detection & Grounding in VLMs20240