Analysing Chain of Thought Dynamics: Active Guidance or Unfaithful Post-hoc Rationalisation?

Recent work has demonstrated that Chain-of-Thought (CoT) often yields limited gains for soft-reasoning problems such as analytical and commonsense reasoning. CoT can also be unfaithful to a model's actual reasoning. We investigate the dynamics and faithfulness of CoT in soft-reasoning tasks across instruction-tuned, reasoning and reasoning-distilled models. Our findings reveal differences in how these models rely on CoT, and show that CoT influence and faithfulness are not always aligned.

It's All About In-Context Learning! Teaching Extremely Low-Resource Languages to LLMs

Extremely low-resource languages, especially those written in rare scripts, as shown in Figure 1, remain largely unsupported by large language models (LLMs). This is due in part to compounding factors such as the lack of training data. This paper delivers the first comprehensive analysis of whether LLMs can acquire such languages purely via in-context learning (ICL), with or without auxiliary alignment signals, and how these methods compare to parameter-efficient fine-tuning (PEFT). We systematically evaluate 20 under-represented languages across three state-of-the-art multilingual LLMs. Our findings highlight the limitation of PEFT when both language and its script are extremely under-represented by the LLM. In contrast, zero-shot ICL with language alignment is impressively effective on extremely low-resource languages, while few-shot ICL or PEFT is more beneficial for languages relatively better represented by LLMs. For LLM practitioners working on extremely low-resource languages, we summarise guidelines grounded by our results on adapting LLMs to low-resource languages, e.g., avoiding fine-tuning a multilingual model on languages of unseen scripts.

Bridging the Gap: In-Context Learning for Modeling Human Disagreement

Large Language Models (LLMs) have shown strong performance on NLP classification tasks. However, they typically rely on aggregated labels-often via majority voting-which can obscure the human disagreement inherent in subjective annotations. This study examines whether LLMs can capture multiple perspectives and reflect annotator disagreement in subjective tasks such as hate speech and offensive language detection. We use in-context learning (ICL) in zero-shot and few-shot settings, evaluating four open-source LLMs across three label modeling strategies: aggregated hard labels, and disaggregated hard and soft labels. In few-shot prompting, we assess demonstration selection methods based on textual similarity (BM25, PLM-based), annotation disagreement (entropy), a combined ranking, and example ordering strategies (random vs. curriculum-based). Results show that multi-perspective generation is viable in zero-shot settings, while few-shot setups often fail to capture the full spectrum of human judgments. Prompt design and demonstration selection notably affect performance, though example ordering has limited impact. These findings highlight the challenges of modeling subjectivity with LLMs and the importance of building more perspective-aware, socially intelligent models.

Tracing and Reversing Rank-One Model Edits

Knowledge editing methods (KEs) are a cost-effective way to update the factual content of large language models (LLMs), but they pose a dual-use risk. While KEs are beneficial for updating outdated or incorrect information, they can be exploited maliciously to implant misinformation or bias. In order to defend against these types of malicious manipulation, we need robust techniques that can reliably detect, interpret, and mitigate adversarial edits. This work investigates the traceability and reversibility of knowledge edits, focusing on the widely used Rank-One Model Editing (ROME) method. We first show that ROME introduces distinctive distributional patterns in the edited weight matrices, which can serve as effective signals for locating the edited weights. Second, we show that these altered weights can reliably be used to predict the edited factual relation, enabling partial reconstruction of the modified fact. Building on this, we propose a method to infer the edited object entity directly from the modified weights, without access to the editing prompt, achieving over 95% accuracy. Finally, we demonstrate that ROME edits can be reversed, recovering the model's original outputs with $\geq$ 80% accuracy. Our findings highlight the feasibility of detecting, tracing, and reversing edits based on the edited weights, offering a robust framework for safeguarding LLMs against adversarial manipulations.

SCRum-9: Multilingual Stance Classification over Rumours on Social Media

We introduce SCRum-9, a multilingual dataset for Rumour Stance Classification, containing 7,516 tweet-reply pairs from X. SCRum-9 goes beyond existing stance classification datasets by covering more languages (9), linking examples to more fact-checked claims (2.1k), and including complex annotations from multiple annotators to account for intra- and inter-annotator variability. Annotations were made by at least three native speakers per language, totalling around 405 hours of annotation and 8,150 dollars in compensation. Experiments on SCRum-9 show that it is a challenging benchmark for both state-of-the-art LLMs (e.g. Deepseek) as well as fine-tuned pre-trained models, motivating future work in this area.

Mitigating Content Effects on Reasoning in Language Models through Fine-Grained Activation Steering

Large language models (LLMs) frequently demonstrate reasoning limitations, often conflating content plausibility (i.e., material inference) with logical validity (i.e., formal inference). This can result in biased inferences, where plausible arguments are incorrectly deemed logically valid or vice versa. Mitigating this limitation is critical, as it undermines the trustworthiness and generalizability of LLMs in applications that demand rigorous logical consistency. This paper investigates the problem of mitigating content biases on formal reasoning through activation steering. Specifically, we curate a controlled syllogistic reasoning dataset to disentangle formal validity from content plausibility. After localising the layers responsible for formal and material inference, we investigate contrastive activation steering methods for test-time interventions. An extensive empirical analysis on different LLMs reveals that contrastive steering consistently supports linear control over content biases. However, we observe that a static approach is insufficient for improving all the tested models. We then leverage the possibility to control content effects by dynamically determining the value of the steering parameters via fine-grained conditional methods. We found that conditional steering is effective on unresponsive models, achieving up to 15% absolute improvement in formal reasoning accuracy with a newly introduced kNN-based method (K-CAST). Finally, additional experiments reveal that steering for content effects is robust to prompt variations, incurs minimal side effects on language modeling capabilities, and can partially generalize to out-of-distribution reasoning tasks. Practically, this paper demonstrates that activation-level interventions can offer a scalable strategy for enhancing the robustness of LLMs, contributing towards more systematic and unbiased formal reasoning.

Position: Editing Large Language Models Poses Serious Safety Risks

Large Language Models (LLMs) contain large amounts of facts about the world. These facts can become outdated over time, which has led to the development of knowledge editing methods (KEs) that can change specific facts in LLMs with limited side effects. This position paper argues that editing LLMs poses serious safety risks that have been largely overlooked. First, we note the fact that KEs are widely available, computationally inexpensive, highly performant, and stealthy makes them an attractive tool for malicious actors. Second, we discuss malicious use cases of KEs, showing how KEs can be easily adapted for a variety of malicious purposes. Third, we highlight vulnerabilities in the AI ecosystem that allow unrestricted uploading and downloading of updated models without verification. Fourth, we argue that a lack of social and institutional awareness exacerbates this risk, and discuss the implications for different stakeholders. We call on the community to (i) research tamper-resistant models and countermeasures against malicious model editing, and (ii) actively engage in securing the AI ecosystem.

Do LLMs Provide Consistent Answers to Health-Related Questions across Languages?

Equitable access to reliable health information is vital for public health, but the quality of online health resources varies by language, raising concerns about inconsistencies in Large Language Models (LLMs) for healthcare. In this study, we examine the consistency of responses provided by LLMs to health-related questions across English, German, Turkish, and Chinese. We largely expand the HealthFC dataset by categorizing health-related questions by disease type and broadening its multilingual scope with Turkish and Chinese translations. We reveal significant inconsistencies in responses that could spread healthcare misinformation. Our main contributions are 1) a multilingual health-related inquiry dataset with meta-information on disease categories, and 2) a novel prompt-based evaluation workflow that enables sub-dimensional comparisons between two languages through parsing. Our findings highlight key challenges in deploying LLM-based tools in multilingual contexts and emphasize the need for improved cross-lingual alignment to ensure accurate and equitable healthcare information.

Implicit Priors Editing in Stable Diffusion via Targeted Token Adjustment

Implicit assumptions and priors are often necessary in text-to-image generation tasks, especially when textual prompts lack sufficient context. However, these assumptions can sometimes reflect outdated concepts, inaccuracies, or societal bias embedded in the training data. We present Embedding-only Editing (Embedit), a method designed to efficiently adjust implict assumptions and priors in the model without affecting its interpretation of unrelated objects or overall performance. Given a "source" prompt (e.g., "rose") that elicits an implicit assumption (e.g., rose is red) and a "destination" prompt that specifies the desired attribute (e.g., "blue rose"), Embedit fine-tunes only the word token embedding (WTE) of the target object ("rose") to optimize the last hidden state of text encoder in Stable Diffusion, a SOTA text-to-image model. This targeted adjustment prevents unintended effects on other objects in the model's knowledge base, as the WTEs for unrelated objects and the model weights remain unchanged. Consequently, when a prompt does not contain the edited object, all representations, and the model outputs are identical to those of the original, unedited model. Our method is highly efficient, modifying only 768 parameters for Stable Diffusion 1.4 and 2048 for XL in a single edit, matching the WTE dimension of each respective model. This minimal scope, combined with rapid execution, makes Embedit highly practical for real-world applications. Additionally, changes are easily reversible by restoring the original WTE layers. Our experimental results demonstrate that Embedit consistently outperforms previous methods across various models, tasks, and editing scenarios (both single and sequential multiple edits), achieving at least a 6.01% improvement (from 87.17% to 93.18%).

ScImage: How Good Are Multimodal Large Language Models at Scientific Text-to-Image Generation?

Multimodal large language models (LLMs) have demonstrated impressive capabilities in generating high-quality images from textual instructions. However, their performance in generating scientific images--a critical application for accelerating scientific progress--remains underexplored. In this work, we address this gap by introducing ScImage, a benchmark designed to evaluate the multimodal capabilities of LLMs in generating scientific images from textual descriptions. ScImage assesses three key dimensions of understanding: spatial, numeric, and attribute comprehension, as well as their combinations, focusing on the relationships between scientific objects (e.g., squares, circles). We evaluate five models, GPT-4o, Llama, AutomaTikZ, Dall-E, and StableDiffusion, using two modes of output generation: code-based outputs (Python, TikZ) and direct raster image generation. Additionally, we examine four different input languages: English, German, Farsi, and Chinese. Our evaluation, conducted with 11 scientists across three criteria (correctness, relevance, and scientific accuracy), reveals that while GPT-4o produces outputs of decent quality for simpler prompts involving individual dimensions such as spatial, numeric, or attribute understanding in isolation, all models face challenges in this task, especially for more complex prompts.