Understanding biases and stereotypes encoded in the weights of Large Language Models (LLMs) is crucial for developing effective mitigation strategies. However, biased behavior is often subtle and non-trivial to isolate, even when deliberately elicited, making systematic analysis and debiasing particularly challenging. To address this, we introduce a simple, cost-effective, and generalizable framework BiasGym for reliably injecting, analyzing, and mitigating conceptual associations of biases within LLMs. BiasGym consists of two modules: Inject, which injects specific biases into the model via token-based fine-tuning while keeping the model frozen, followed by two debiasing methods that leverage these injected signals to identify and reliably suppress (Scope) or Steer the components responsible for biased behavior. Our framework enables consistent bias elicitation for better localization of bias conceptual association in the model space, supports targeted debiasing without degrading performance on downstream tasks, and generalizes to biases unseen during fine-tuning. We demonstrate the effectiveness of our proposed framework in reducing real-world stereotypes (e.g., people from Italy being ‘reckless drivers’), showing its utility for both safety interventions and interpretability research.

Natural Language Explanations (NLEs) describe how Large Language Models (LLMs) make decisions by drawing on external Context Knowledge (CK) and Parametric Knowledge (PK). Understanding the interaction between these sources is key to assessing NLE grounding, yet these dynamics remain underexplored. Prior work has largely focused on i) single-step generation and ii) modeled PK–CK interaction as a binary choice within a rank-1 subspace. This approach overlooks richer interactions and how they unfold over longer generations, such as complementary or supportive knowledge. We propose a novel rank-2 projection subspace that disentangles PK and CK contributions more accurately and use it for the first multi-step analysis of knowledge interactions across longer NLE sequences. Experiments across four QA datasets and three open-weight LLMs demonstrate that rank-1 subspaces struggle to represent diverse interactions, whereas our rank-2 formulation captures them effectively, highlighting PK alignment for supportive interactions and CK alignment for conflicting ones. Our multi-step analysis reveals that hallucinated generations exhibit strong alignment with the PK direction, though context-faithful generations maintain a more balanced alignment between PK and CK.

Context utilization, the ability of Language Models (LMs) to incorporate relevant information from the provided context when generating responses, remains largely opaque to users, who cannot determine whether models draw from parametric memory or provided context, nor identify which specific context pieces inform the response. Highlight explanations (HEs) offer a natural solution as they can point to the exact context pieces and tokens that influenced model outputs. However, no existing work evaluates their effectiveness in accurately explaining context utilization. We address this gap by introducing the first gold standard HE evaluation framework for context attribution, using controlled test cases with known ground-truth context usage, thereby avoiding the limitations of existing indirect proxy evaluations. To demonstrate the framework’s broad applicability, we evaluate four HE methods—three established techniques and MechLight, a mechanistic interpretability approach we adapt for this task—across four context scenarios, four datasets, and five LMs. Overall, we find that MechLight performs best across all context scenarios. However, our findings reveal systematic failures in all methods: Explanation accuracy degrades significantly with context length, and all methods exhibit strong positional biases in multi-document settings. Surprisingly, widely used gradient-based methods provide little value for understanding context usage. These results challenge HEs’ utility in retrieval-augmented generation, factual verification, and other applications. Our framework provides the foundation for developing accurate context attribution methods.