Multimodal Model Diffing for Feature Discovery and Control | Hunar Batra et al. | ResearchPod

Transcript: Multimodal Model Diffing for Feature Discovery and Control

Alex: Welcome to another episode of ResearchPod. Today, we're looking at a paper that asks a deceptively simple question: when an AI model that handles both images and text makes a mistake, how do you find and fix the exact internal part responsible — without rebuilding the whole system from scratch?

Sam: That sounds like a genuinely hard problem. Why is the usual fix so costly?

Alex: Because these models are enormous. Retraining one from scratch to correct a single flaw takes weeks of computing time and enormous expense — and it risks breaking skills the model already had. It's like tearing down a wall to fix one bad pipe, and accidentally knocking out the electricity in the process.

Sam: So the researchers are looking for a more targeted approach. What's their starting point?

Alex: They start by trying to understand what actually changed inside the model when it learned to handle images. They call this "model diffing." Think of it like comparing two blueprints of the same building — one from before a renovation, one from after — to see exactly what was added or moved. Here, the "before" is a text-only AI, and the "after" is the same model once it's been trained to understand images as well.

Sam: So rather than just looking at what the model outputs, they're comparing its internal activity at each stage of processing.

Alex: Right. To do that, they attach small helper networks to the model's internal states. These helpers are trained to compress the model's activity into a set of clean, distinct directions — like taking a massive, chaotic library and sorting every book onto labelled shelves, so you can actually see what's where.

Sam: And once things are sorted, you can spot which shelves got reorganised when the visual wing was added.

Alex: Precisely. But there's a subtlety worth flagging. The model processes both text and images, so how do you make sure you're only identifying features that genuinely changed because of visual training — and not just background noise?

Sam: How do they handle that?

Alex: Two ways. First, they train the helper networks using text tokens only. That keeps the internal measuring stick anchored to the original language model, so any shift they detect is specifically caused by the visual training — not by general drift. Like calibrating your ruler before you measure anything.

Sam: That makes sense. What's the second step?

Alex: They use two signals to score each internal feature. One measures how strongly a feature activates when an image is present. The other measures how much that feature's internal wiring actually shifted during training. A feature that scores high on both has genuinely been repurposed to handle visual input — rather than just reusing old text habits.

Multimodal Model Diffing for Feature Discovery and Control

Hunar Batra, Lachin Naghashyar, Ashkan Khakzar, Philip Torr, Christian Schroeder de Witt, Constantin Venhoff, Ronald Clark

Aug 10, 2026·5:01·arxiv·Original Paper

Multimodal Model Diffing (MMDiff)Sparse Autoencoders (SAEs)Feature-level ControlContrastive Firing AnalysisMMDiffMultimodal Model Diffing

Listen

5 min

Briefing

Research Question and Approach

Multimodal Large Language Models (MLLMs) possess impressive visual capabilities, but their internal mechanisms remain largely opaque. Researchers struggle to identify which internal features drive specific behaviors, making it difficult to audit or steer these models without expensive retraining. This paper introduces MMDiff, a framework that uses sparse autoencoders (SAEs) to create a feature-level interface for MLLMs. By comparing SAEs trained on a base language model against those trained on the multimodal-adapted version, the authors isolate features that were specifically reshaped or repurposed by multimodal training. The pipeline involves three key steps: isolating visually responsive features, identifying task-specific subsets through contrastive firing analysis, and applying causal interventions to remove or steer these features.

Main Findings

MMDiff successfully identifies sparse, causally specific features across three MLLM families: LLaVA-MORE, PaliGemma 2, and InternVL3.5. The authors demonstrate that these features are not just interpretability artifacts but functional control handles. Causal removal of these features selectively degrades target behaviors—reducing spatial reasoning accuracy by 12%, OCR accuracy by 17%, and multimodal safety attack success rates by 24%—without negatively impacting general visual question answering performance. Furthermore, the authors introduce MMDiff-CAA, a steering method that combines backbone activation steering with the injection of specific SAE decoder directions. This approach improves spatial and OCR accuracy by an average of 3.6% and 1.8%, respectively, outperforming standard single-layer steering baselines.

Why It Matters

This work provides a bridge between interpretability and model control. By demonstrating that multimodal SAEs can serve as mechanisms for auditing and steering, the authors offer a path toward safer and more reliable MLLMs. The ability to selectively suppress unsafe behaviors or enhance specific capabilities like spatial reasoning without retraining the entire model is a significant step forward for model governance and alignment.

Key Terms and Definitions

Model Diffing — A technique comparing internal representations or features across different training stages to isolate changes induced by specific learning objectives.
Sparse Autoencoder (SAE) — A neural network architecture used to decompose high-dimensional hidden states into a sparse set of interpretable feature directions.
Contrastive Firing Analysis — A method of identifying features that activate significantly more frequently under a target distribution compared to a baseline distribution.
Causal Removal — An intervention technique that orthogonally projects out a specific feature direction from the model's residual stream to test its necessity for a given behavior.
MMDiff-CAA — A steering strategy that combines multi-layer backbone activation steering with the injection of a specific SAE decoder direction to amplify target model behaviors.

Abstract

Multimodal Large Language Models (MLLMs) exhibit strong visual understanding, yet the internal features that cause these behaviors remain difficult to identify, audit, or control. While applicable to post-hoc inspection, hidden states that are decomposed into interpretable feature directions using sparse autoencoders (SAEs) neither readily isolate which features are changed by multimodal training, nor are they directly useful for targeted control. We introduce MMDiff, a multimodal model-diffing framework that trains multimodal SAEs and turns them into feature-level interfaces for discovering and controlling multimodal behavior. MMDiff supports three uses: (i) feature isolation, by diffing a base-LM SAE against its multimodal-adapted counterpart to identify features altered by multimodal training; (ii) task-specific feature detection, via per-token contrastive firing analysis that isolates causal features; and (iii) feature-level control, by causally removing or steering the discovered feature directions. We train multimodal SAEs for three MLLM families, LLaVA-MORE, PaliGemma 2, and InternVL3.5, and evaluate on visual-spatial understanding, multimodal safety, and OCR. MMDiff discovers sparse, causally specific features whose removal selectively degrades target behaviors by an average of 12% on spatial tasks and 17% on OCR, and reduces attack success rate by 24% on multimodal safety attacks, with no impact on VQA performance. Steering these features improves spatial and OCR accuracy by +3.6% and +1.8% on average over a standard single-layer steering baseline. These results show that multimodal SAEs can serve not only as interpretability tools, but as mechanisms for auditing, steering, and controlling MLLMs behavior toward safer and more capable generations.

Multimodal Model Diffing for Feature Discovery and Control: Research Paper Podcast