Multimodal Model Diffing for Feature Discovery and Control

Hunar Batra, Lachin Naghashyar, Ashkan Khakzar, Philip Torr, Christian Schroeder de Witt et al.|August 10, 2026arXiv

Key Takeaway

Sparse autoencoders can turn multimodal model internals into controllable feature interfaces: you can identify what changed during multimodal training, find features causing specific behaviors, and steer or remove them to improve safety or task performance.

Summary

This paper introduces MMDiff, a framework that uses sparse autoencoders to decompose multimodal language models into interpretable features.

multimodal safety

Key Terms

sparse-autoencoder model-diffing activation-patching mechanistic-interpretability steering-vectors