Output Vector Editing for Memorization Mitigation in Large Language Models

Large language models memorize and reproduce sequences from their training data, creating privacy, copyright, and security risks. Existing neuron-level mitigation methods equate editing with zeroing out neuron activations, but the activation only controls whether a neuron engages; the output vector is what writes to the residual stream and, through superposition, encodes multiple features. We propose output vector editing, a constrained-optimization weight edit that locates a small set of MLP neurons responsible for a memorized continuation and minimally modifies their output vectors to introduce a distractor in vocabulary space, redirecting their residual-stream contributions while leaving activations unchanged. Evaluating on four models from 360M to 7B parameters (SmolLM-360M, OLMo-1B, OLMo-7B, Llama2-7B), we center on OLMo-7B (whose open weights and pretraining corpus enable systematic mining) and mine 6831 memorized sequences, achieving up to 87.9% suppression. The 2.7$\times$ gap over zero ablation on the same located neurons shows the suppression comes from the output-vector edit, not localization alone. Four edit modes span a spectrum from aggressive suppression to minimal redirection; in ensemble they cover 96.5% of memorized sequences, while our recommended single-mode configuration reaches 81.5% with no catastrophic locality failures. We further identify a mechanistic boundary at ${\sim}14%$ of sequences unreachable by MLP-only editing; while these failures are not attention-driven overall, ablating the top contributing attention heads recovers 60--64% of them, with stronger recovery on continuations that copy tokens from the prefix, positioning attention as a complementary fallback rather than a primary mechanism. Edit mode ordering and the success-locality trade-off transfer across all four models, with success rates scaling with model size rather than family.

Emergence of Context Characteristics Sensitivity in Large Language Models

During instruction fine-tuning (IFT), large language models (LLMs) learn to follow instructions by using the provided context to answer a query. While prior work has studied how context characteristics correlate with context usage by the LLM, this analysis has been limited to inference time, leaving open how these relationships are acquired in the first place. Here, we measure how models' sensitivity to such characteristics shifts across successive IFT stages: supervised fine-tuning (SFT), direct preference optimization (DPO), and reinforcement learning with verifiable rewards (RLVR). Experiments across four models and three datasets show that SFT makes models more likely to use contexts that are easy to understand, such as containing high length, context-query similarity, and fluency. Post-SFT dynamics may either reinforce or resolve these preferences depending on the training dataset. Our findings reveal that context usage is actively reshaped at each IFT stage, and designing a balanced IFT dataset is important in ensuring robust context utilization of instruction-tuned models.

Whose Norms? Disentangling Cultural and Personal Alignment in Large Language Models

Large language models are increasingly used for social decision-making situations that require balancing cultural norms with personal preferences. For example, a user preferring honesty might ask whether to correct a coworker publicly when local norms favor indirect feedback. Yet existing research studies cultural alignment and personalization largely separately. We introduce PACT, the Personal-Preference and Cultural-Norm Trade-off framework, which evaluates whether models choose to follow a cultural norm or allow personal preferences. We find that LLMs vary in how rigidly they enforce cultural norms, with behavior shifted more by country context (7.8%) than age (1%) and gender (0.7%) and shifting non-uniformly after instruction tuning. Furthermore, our five-country human study on PACT shows that culture-following in humans is mainly driven by scenario country, with the lowest agreement when participants judge their own cultural contexts, showing within-culture pluralism. Finally, human-LLM alignment experiments show that models can match majority choices, but fail to capture response distributions and uncertainty (with best correlations reaching only 0.24). Together, these findings motivate alignment evaluations that go beyond majority to capture cultural pluralism and disagreement in social judgment.

Not What, But How: A Framework for Auditing LLM Responses across Positioning, Generalization, Anthromorphism, and Maxims

Large language models (LLMs) are being increasingly used to answer subjective, information-seeking questions, where users are sensitive to how responses are communicated, not just whether the answers are correct. Existing LLM evaluations for subjective cultural queries largely focus on factual correctness, ignoring how the response is framed. To this end, we introduce FRANZ, an automated FRAmework for respoNse characteriZation to conduct communicative audit of LLM responses along four dimensions: cultural positioning, use of generalizing language, anthropomorphic cues, and adherence to conversational maxims. To enable this evaluation, we contribute SQUARE - a corpus of 376k subjective questions sourced from 57 subreddits, and mapped to 7 countries and 19 question categories. We demonstrate FRANZ's applicability by scoring responses from three open-weight LLMs. We observe that LLMs show statistically significant differences in the frequency with which they employ each response characteristic. Unlike single-dimensional audits, FRANZ reveals that insider positioning and anthropomorphism are positively coupled, with the degree of coupling varying by country, providing a diagnostic lens for identifying framing divergences.

Not What, But How: A Communicative Audit of LLM Response Framing

Large language models (LLMs) are being increasingly used to answer subjective, information-seeking questions, where users are sensitive to how responses are communicated, not just whether the answers are correct. Existing LLM evaluations for subjective cultural queries largely focus on factual correctness, ignoring how the response is framed. To this end, we introduce FRANZ, an automated FRAmework for respoNse characteriZation to conduct communicative audit of LLM responses along four dimensions: cultural positioning, use of generalizing language, anthropomorphic cues, and adherence to conversational maxims. To enable this evaluation, we contribute SQUARE - a corpus of 376k subjective questions sourced from 57 subreddits, and mapped to 7 countries and 19 question categories. We demonstrate FRANZ's applicability by scoring responses from three open-weight LLMs. We observe that LLMs show statistically significant differences in the frequency with which they employ each response characteristic. Unlike single-dimensional audits, FRANZ reveals that insider positioning and anthropomorphism are positively coupled, with the degree of coupling varying by country, providing a diagnostic lens for identifying framing divergences.

Investigating the Interplay between Contextual and Parametric Chain-of-Thought Faithfulness under Optimization

Chain-of-Thought (CoT) faithfulness, i.e., whether CoTs genuinely reflect large language models' (LLM) underlying behavior, is typically evaluated under two disjoint paradigms: contextual faithfulness, measured by perturbing the input or CoT trace, and parametric faithfulness, assessed by intervening on a model's parametric knowledge. Yet prior work compares them only descriptively. We fill this gap by proposing FaithMate, a unified preference-alignment interface for optimizing models towards either faithfulness paradigm. It enables us to investigate the interplay between the two paradigms, examining whether and to what extent faithfulness gains generalize within and across paradigms. Across three models, two datasets, and six faithfulness metrics, we find that the two paradigms are positively coupled, yet asymmetric: optimizing towards parametric faithfulness yields consistent gains across both paradigms, whereas the contextual counterpart delivers more variable gains. Within the contextual paradigm, faithfulness gains on one metric do not consistently transfer to others, implying that existing contextual metrics capture disjoint facets of faithfulness and exposing inherent trade-offs. These findings imply that CoT faithfulness is not a monolithic objective and therefore requires multifaceted optimization and evaluation.

Do We Still Need Humans in the Loop? Comparing Human and LLM Annotation in Active Learning for Hostility Detection

Instruction-tuned LLMs can annotate thousands of instances from a short prompt at negligible cost. This raises two questions for active learning (AL): can LLM labels replace human labels within the AL loop, and does AL remain necessary when entire corpora can be labelled at once? We investigate both questions on a new dataset of 277,902 German political TikTok comments (25,974 LLM-labelled, 5,000 human-annotated), comparing seven annotation strategies across four encoders to detect anti-immigrant hostility. A classifier trained on 25,974 GPT-5.2 labels (\$43) achieves comparable F1-Macro to one trained on 3,800 human annotations (\$316). Active learning offers little advantage over random sampling in our pre-enriched pool and delivers lower F1 than full LLM annotation at the same cost. However, comparable aggregate F1 masks a systematic difference in error structure: LLM-trained classifiers over-predict the positive class relative to the human gold standard. This divergence concentrates in topically ambiguous discussions where the distinction between anti-immigrant hostility and policy critique is most subtle, suggesting that annotation strategy should be guided not by aggregate F1 alone but by the error profile acceptable for the target application.

Understanding the Interplay between LLMs' Utilisation of Parametric and Contextual Knowledge: A keynote at ECIR 2025

Language Models (LMs) acquire parametric knowledge from their training process, embedding it within their weights. The increasing scalability of LMs, however, poses significant challenges for understanding a model's inner workings and further for updating or correcting this embedded knowledge without the significant cost of retraining. Moreover, when using these language models for knowledge-intensive language understanding tasks, LMs have to integrate relevant context, mitigating their inherent weaknesses, such as incomplete or outdated knowledge. Nevertheless, studies indicate that LMs often ignore the provided context as it can be in conflict with the pre-existing LM's memory learned during pre-training. Conflicting knowledge can also already be present in the LM's parameters, termed intra-memory conflict. This underscores the importance of understanding the interplay between how a language model uses its parametric knowledge and the retrieved contextual knowledge. In this talk, I will aim to shed light on this important issue by presenting our research on evaluating the knowledge present in LMs, diagnostic tests that can reveal knowledge conflicts, as well as on understanding the characteristics of successfully used contextual knowledge.

Interpretable Debiasing of Vision-Language Models for Social Fairness

The rapid advancement of Vision-Language models (VLMs) has raised growing concerns that their black-box reasoning processes could lead to unintended forms of social bias. Current debiasing approaches focus on mitigating surface-level bias signals through post-hoc learning or test-time algorithms, while leaving the internal dynamics of the model largely unexplored. In this work, we introduce an interpretable, model-agnostic bias mitigation framework, DeBiasLens, that localizes social attribute neurons in VLMs through sparse autoencoders (SAEs) applied to multimodal encoders. Building upon the disentanglement ability of SAEs, we train them on facial image or caption datasets without corresponding social attribute labels to uncover neurons highly responsive to specific demographics, including those that are underrepresented. By selectively deactivating the social neurons most strongly tied to bias for each group, we effectively mitigate socially biased behaviors of VLMs without degrading their semantic knowledge. Our research lays the groundwork for future auditing tools, prioritizing social fairness in emerging real-world AI systems.

Analysing Differences in Persuasive Language in LLM-Generated Text: Uncovering Stereotypical Gender Patterns

Large language models (LLMs) are increasingly used for everyday communication tasks, including drafting interpersonal messages intended to influence and persuade. Prior work has shown that LLMs can successfully persuade humans and amplify persuasive language. It is therefore essential to understand how user instructions affect the generation of persuasive language, and to understand whether the generated persuasive language differs, for example, when targeting different groups. In this work, we propose a framework for evaluating how persuasive language generation is affected by recipient gender, sender intent, or output language. We evaluate 13 LLMs and 16 languages using pairwise prompt instructions. We evaluate model responses on 19 categories of persuasive language using an LLM-as-judge setup grounded in social psychology and communication science. Our results reveal significant gender differences in the persuasive language generated across all models. These patterns reflect biases consistent with gender-stereotypical linguistic tendencies documented in social psychology and sociolinguistics.