Cross-Architecture Model Diffing with Crosscoders: Unsupervised Discovery of Differences Between LLMs
This work addresses the need for model diffing across novel architectures to uncover safety-critical behaviors in AI models, though it is incremental as it builds on existing crosscoder methods.
The paper tackled the problem of comparing AI models with different architectures by applying crosscoders for cross-architecture model diffing and introducing Dedicated Feature Crosscoders to better isolate unique features, resulting in unsupervised discovery of features such as Chinese Communist Party alignment, American exceptionalism, and a copyright refusal mechanism in various LLMs.
Model diffing, the process of comparing models' internal representations to identify their differences, is a promising approach for uncovering safety-critical behaviors in new models. However, its application has so far been primarily focused on comparing a base model with its finetune. Since new LLM releases are often novel architectures, cross-architecture methods are essential to make model diffing widely applicable. Crosscoders are one solution capable of cross-architecture model diffing but have only ever been applied to base vs finetune comparisons. We provide the first application of crosscoders to cross-architecture model diffing and introduce Dedicated Feature Crosscoders (DFCs), an architectural modification designed to better isolate features unique to one model. Using this technique, we find in an unsupervised fashion features including Chinese Communist Party alignment in Qwen3-8B and Deepseek-R1-0528-Qwen3-8B, American exceptionalism in Llama3.1-8B-Instruct, and a copyright refusal mechanism in GPT-OSS-20B. Together, our results work towards establishing cross-architecture crosscoder model diffing as an effective method for identifying meaningful behavioral differences between AI models.