How do language models use information provided as context when generating a response?

Removing specific data influence from large language models (LLMs) remains challenging, as retraining is costly and existing approximate unlearning methods are often unstable. The challenge is exacerbated when the forget set is small or imbalanced. We introduce RapidUn, an influence-driven and parameter-efficient unlearning framework. It first estimates per-sample influence through a fast estimation module, then maps these scores into adaptive update weights that guide selective parameter updates -- forgetting harmful behavior while retaining general knowledge. On Mistral-7B and Llama-3-8B across Dolly-15k and Alpaca-57k, RapidUn achieves up to 100 times higher efficiency than full retraining and consistently outperforms Fisher, GA, and LoReUn on both in-distribution and out-of-distribution forgetting. These results establish influence-guided parameter reweighting as a scalable and interpretable paradigm for LLM unlearning.

Recent studies suggest that the representations learned by large language models (LLMs) are partially aligned to those of the human brain. However, whether and why this alignment score arises from a similar sequence of computations remains elusive. In this study, we explore this question by examining temporally-resolved brain signals of participants listening to 10 hours of an audiobook. We study these neural dynamics jointly with a benchmark encompassing 22 LLMs varying in size and architecture type. Our analyses confirm that LLMs and the brain generate representations in a similar order: specifically, activations in the initial layers of LLMs tend to best align with early brain responses, while the deeper layers of LLMs tend to best align with later brain responses. This brain-LLM alignment is consistent across transformers and recurrent architectures. However, its emergence depends on both model size and context length. Overall, this study sheds light on the sequential nature of computations and the factors underlying the partial convergence between biological and artificial neural networks.

As teachers increasingly turn to GenAI in their educational practice, we need robust methods to benchmark large language models (LLMs) for pedagogical purposes. This article presents an embedding-based benchmarking framework to detect bias in LLMs in the context of formative feedback. Using 600 authentic student essays from the AES 2.0 corpus, we constructed controlled counterfactuals along two dimensions: (i) implicit cues via lexicon-based swaps of gendered terms within essays, and (ii) explicit cues via gendered author background in the prompt. We investigated six representative LLMs (i.e. GPT-5 mini, GPT-4o mini, DeepSeek-R1, DeepSeek-R1-Qwen, Gemini 2.5 Pro, Llama-3-8B). We first quantified the response divergence with cosine and Euclidean distances over sentence embeddings, then assessed significance via permutation tests, and finally, visualised structure using dimensionality reduction. In all models, implicit manipulations reliably induced larger semantic shifts for male-female counterfactuals than for female-male. Only the GPT and Llama models showed sensitivity to explicit gender cues. These findings show that even state-of-the-art LLMs exhibit asymmetric semantic responses to gender substitutions, suggesting persistent gender biases in feedback they provide learners. Qualitative analyses further revealed consistent linguistic differences (e.g., more autonomy-supportive feedback under male cues vs. more controlling feedback under female cues). We discuss implications for fairness auditing of pedagogical GenAI, propose reporting standards for counterfactual evaluation in learning analytics, and outline practical guidance for prompt design and deployment to safeguard equitable feedback.

With the growing computational demands of large language models (LLMs), efficient inference has become increasingly critical for practical deployment. Depth pruning has emerged as a promising approach for reducing the computational costs of large language models by removing transformer layers. However, existing methods typically rely on fixed block masks, which can lead to suboptimal performance across different tasks and inputs. In this paper, we propose IG-Pruning, a novel input-aware block-wise pruning method that dynamically selects layer masks at inference time. Our approach consists of two stages: (1) Discovering diverse mask candidates through semantic clustering and L0 optimization, and (2) Implementing efficient dynamic pruning without the need for extensive training. Experimental results demonstrate that our method consistently outperforms state-of-the-art static depth pruning methods, making it particularly suitable for resource-constrained deployment scenarios.

Large Language Models (LLMs) demonstrate strong performance but often lack interpretable reasoning. This paper introduces the Multi-Agent Collaboration Framework for Diverse Thinking Modes (DiMo), which enhances both performance and interpretability by simulating a structured debate among four specialized LLM agents. Each agent embodies a distinct reasoning paradigm, allowing the framework to collaboratively explore diverse cognitive approaches. Through iterative debate, agents challenge and refine initial responses, yielding more robust conclusions and an explicit, auditable reasoning chain. Across six benchmarks and under a unified open-source setup, DiMo improves accuracy over widely used single-model and debate baselines, with the largest gains on math. We position DiMo as a semantics-aware, Web-native multi-agent framework: it models human-machine intelligence with LLM agents that produce semantically typed, URL-annotated evidence chains for explanations and user-friendly interactions. Although our experiments use standard reasoning benchmarks, the framework is designed to be instantiated over Web corpora and knowledge graphs, combining retrieval-augmented reasoning with structured justifications that downstream systems can inspect and reuse.

Although Large Language Models (LLMs) perform well in general fields, they exhibit a confidence distortion problem on multi-choice question-answering (MCQA), particularly as the number of answer choices increases. Specifically, on MCQA with many choices, LLMs suffer from under-confidence in correct predictions and over-confidence in incorrect ones, leading to a substantially degraded performance. To solve this problem, we propose Self-ensemble in this work. Our method splits the choices into several groups and ensembles LLM predictions across these groups to reach a final decision. The advantage of Self-ensemble is its plug-and-play nature, where it can be integrated into existing LLM architecture based on a designed attention mask and positional encoding, without requiring labeled datasets for parameter tuning. Experimental results on three LLMs and datasets demonstrate that Self-ensemble comprehensively addresses the confidence distortion problem of LLMs, outperforming standard inference as well as baseline methods.

Recent work suggests that large language models (LLMs) encode factuality signals in their internal representations, such as hidden states, attention weights, or token probabilities, implying that LLMs may "know what they don't know". However, LLMs can also produce factual errors by relying on shortcuts or spurious associations. These error are driven by the same training objective that encourage correct predictions, raising the question of whether internal computations can reliably distinguish between factual and hallucinated outputs. In this work, we conduct a mechanistic analysis of how LLMs internally process factual queries by comparing two types of hallucinations based on their reliance on subject information. We find that when hallucinations are associated with subject knowledge, LLMs employ the same internal recall process as for correct responses, leading to overlapping and indistinguishable hidden-state geometries. In contrast, hallucinations detached from subject knowledge produce distinct, clustered representations that make them detectable. These findings reveal a fundamental limitation: LLMs do not encode truthfulness in their internal states but only patterns of knowledge recall, demonstrating that "LLMs don't really know what they don't know".

Retrieval-augmented generation (RAG) has become a cornerstone of contemporary NLP, enhancing large language models (LLMs) by allowing them to access richer factual contexts through in-context retrieval. While effective in monolingual settings, especially in English, its use in multilingual tasks remains unexplored. This paper investigates the effectiveness of RAG across multiple languages by proposing novel approaches for multilingual open-domain question-answering. We evaluate the performance of various multilingual RAG strategies, including question-translation (tRAG), which translates questions into English before retrieval, and Multilingual RAG (MultiRAG), where retrieval occurs directly across multiple languages. Our findings reveal that tRAG, while useful, suffers from limited coverage. In contrast, MultiRAG improves efficiency by enabling multilingual retrieval but introduces inconsistencies due to cross-lingual variations in the retrieved content. To address these issues, we propose Crosslingual RAG (CrossRAG), a method that translates retrieved documents into a common language (e.g., English) before generating the response. Our experiments show that CrossRAG significantly enhances performance on knowledge-intensive tasks, benefiting both high-resource and low-resource languages.