# Transcoders Trace Visual Grounding and Hallucinations in Vision-Language Models

arXiv: https://arxiv.org/abs/2605.22902 · HTML: https://mlsanity.com/paper/2605.22902/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Multimodal reasoning, one of 17 directions in LLM reasoning.
- Builds on 28 of its 38 references that are in the map.
- First posted on arXiv 21 May 2026.

## Placed in

- LLM reasoning / [Multimodal reasoning](https://mlsanity.com/lr/multimodal-reasoning/)
- Multimodal LLMs / vision-language models / [VLM benchmarks and visual reasoning](https://mlsanity.com/mllm/vlm-benchmarks-and-visual-reasoning/)
- Mechanistic interpretability / [Interpreting vision and multimodal models](https://mlsanity.com/mechinterp/interpreting-vision-and-multimodal-models/)

## Builds on (most cited first, 28 of 28)

- [Learning Transferable Visual Models From Natural Language Supervision](https://mlsanity.com/paper/2103.00020/) (2021)
- [Microsoft COCO: Common Objects in Context](https://mlsanity.com/paper/1405.0312/) (2014)
- [Visual Instruction Tuning](https://mlsanity.com/paper/2304.08485/) (2023)
- [A Unified Approach to Interpreting Model Predictions](https://mlsanity.com/paper/1705.07874/) (2017)
- [Sigmoid Loss for Language Image Pre-Training](https://mlsanity.com/paper/2303.15343/) (2023)
- [Qwen3-VL Technical Report](https://mlsanity.com/paper/2511.21631/) (2025)
- [Making the V in VQA Matter: Elevating the Role of Image Understanding in Visual Question Answering](https://mlsanity.com/paper/1612.00837/) (2016)
- [Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Maps](https://mlsanity.com/paper/1312.6034/) (2013)
- [CLEVR: A Diagnostic Dataset for Compositional Language and Elementary Visual Reasoning](https://mlsanity.com/paper/1612.06890/) (2016)
- [Sparse Autoencoders Find Highly Interpretable Features in Language Models](https://mlsanity.com/paper/2309.08600/) (2023)
- [OpenAI GPT-5 System Card](https://mlsanity.com/paper/2601.03267/) (2025)
- [Cross-Task Generalization via Natural Language Crowdsourcing Instructions](https://mlsanity.com/paper/2104.08773/) (2021)
- [Scaling and evaluating sparse autoencoders](https://mlsanity.com/paper/2406.04093/) (2024)
- [Mechanistic Interpretability for AI Safety -- A Review](https://mlsanity.com/paper/2404.14082/) (2024)
- [Persona Vectors: Monitoring and Controlling Character Traits in Language Models](https://mlsanity.com/paper/2507.21509/) (2025)
- [Explainability for Large Language Models: A Survey](https://mlsanity.com/paper/2309.01029/) (2023)
- [Transcoders Find Interpretable LLM Feature Circuits](https://mlsanity.com/paper/2406.11944/) (2024)
- [SAEs Are Good for Steering -- If You Select the Right Features](https://mlsanity.com/paper/2505.20063/) (2025)
- [Sparse Autoencoders Learn Monosemantic Features in Vision-Language Models](https://mlsanity.com/paper/2504.02821/) (2025)
- [Interpreting CLIP with Hierarchical Sparse Autoencoders](https://mlsanity.com/paper/2502.20578/) (2025)
- [Interpretable Steering of Large Language Models with Feature Guided Activation Additions](https://mlsanity.com/paper/2501.09929/) (2025)
- [HaloQuest: A Visual Hallucination Dataset for Advancing Multimodal Reasoning](https://mlsanity.com/paper/2407.15680/) (2024)
- [Interpretable and Testable Vision Features via Sparse Autoencoders](https://mlsanity.com/paper/2502.06755/) (2025)
- [SAE-V: Interpreting Multimodal Models for Enhanced Alignment](https://mlsanity.com/paper/2502.17514/) (2025)
- [Controllable LLM Reasoning via Sparse Autoencoder-Based Steering](https://mlsanity.com/paper/2601.03595/) (2026)
- [When Seeing Overrides Knowing: Disentangling Knowledge Conflicts in Vision-Language Models](https://mlsanity.com/paper/2507.13868/) (2025)
- [Circuit Tracing in Vision-Language Models: Understanding the Internal Mechanisms of Multimodal Thinking](https://mlsanity.com/paper/2602.20330/) (2026)
- [ActivationReasoning: Logical Reasoning in Latent Activation Spaces](https://mlsanity.com/paper/2510.18184/) (2025)

## Abstract (arXiv)

Generative Vision-Language Models (VLMs) perform well on multimodal reasoning, but how visual inputs are transformed to text remains poorly understood. Existing interpretability work on VLMs uses Sparse Autoencoders (SAEs), which decompose static residual representations and miss the functional updates that drive cross-modal interaction. We adopt a function-centric framework based on Transcoders, sparse approximations of MLP sublayers that act as a causal proxy for layer-wise computation. Applied to Gemma 3-4B-IT, the framework decomposes the model into interpretable computational pathways linking image patches to directions in token generation. Transcoder attributions produce stronger and more stable effects on visually grounded tokens under patch ablation than SAE attributions, and align better with semantically relevant image regions. A False Visual Grounding counterfactual analysis confirms that the recovered pathways are specific to vision-language interaction.Finally, we perform a structural analysis of hallucinated generations, by extracting graph-based indicators from circuit traces produced by the transcoders. A logistic classifier over these mechanistic graph features predicts hallucinations at AUC 0.68. These results show that function-centric circuit decomposition yields interpretable and predictive accounts of multimodal computation in VLMs.