Aligned Probing: Relating Toxic Behavior and Model Internals
This provides a novel interpretability framework for understanding and mitigating toxic behavior in language models, with incremental improvements in bridging behavioral and internal perspectives.
The paper introduced aligned probing to link language model outputs with internal representations, analyzing over 20 models to show they encode toxicity information strongly in lower layers, leading to less toxic outputs when this encoding is robust.
We introduce aligned probing, a novel interpretability framework that aligns the behavior of language models (LMs), based on their outputs, and their internal representations (internals). Using this framework, we examine over 20 OLMo, Llama, and Mistral models, bridging behavioral and internal perspectives for toxicity for the first time. Our results show that LMs strongly encode information about the toxicity level of inputs and subsequent outputs, particularly in lower layers. Focusing on how unique LMs differ offers both correlative and causal evidence that they generate less toxic output when strongly encoding information about the input toxicity. We also highlight the heterogeneity of toxicity, as model behavior and internals vary across unique attributes such as Threat. Finally, four case studies analyzing detoxification, multi-prompt evaluations, model quantization, and pre-training dynamics underline the practical impact of aligned probing with further concrete insights. Our findings contribute to a more holistic understanding of LMs, both within and beyond the context of toxicity.