In-Context Learning (ICL) has emerged as an important new paradigm in natural language processing and large language model (LLM) applications. However, the theoretical understanding of the ICL mechanism remains limited. This paper aims to investigate this issue by studying a particular ICL approach, called concept-based ICL (CB-ICL). In particular, we propose theoretical analyses on applying CB-ICL to ICL tasks, which explains why and when the CB-ICL performs well for predicting query labels in prompts with only a few demonstrations. In addition, the proposed theory quantifies the knowledge that can be leveraged by the LLMs to the prompt tasks, and leads to a similarity measure between the prompt demonstrations and the query input, which provides important insights and guidance for model pre-training and prompt engineering in ICL. Moreover, the impact of the prompt demonstration size and the dimension of the LLM embeddings in ICL are also explored based on the proposed theory. Finally, several real-data experiments are conducted to validate the practical usefulness of CB-ICL and the corresponding theory. With the great successes of large language models (LLMs), In-context learning (ICL) has emerged as a new paradigm for natural language processing (NLP) (Brown et al., 2020; Chowdhery et al., 2023; Achiam et al., 2023), where LLMs addresses the requested queries in context prompts with a few demonstrations.

We introduce Drivelology, a unique linguistic phenomenon characterised as "nonsense with depth" - utterances that are syntactically coherent yet pragmatically paradoxical, emotionally loaded, or rhetorically subversive. While such expressions may resemble surface-level nonsense, they encode implicit meaning requiring contextual inference, moral reasoning, or emotional interpretation. We find that current large language models (LLMs), despite excelling at many natural language processing (NLP) tasks, consistently fail to grasp the layered semantics of Drivelological text. To investigate this, we construct a benchmark dataset of over 1,200+ meticulously curated and diverse examples across English, Mandarin, Spanish, French, Japanese, and Korean. Each example underwent careful expert review to verify its Drivelological characteristics, involving multiple rounds of discussion and adjudication to address disagreements. Using this dataset, we evaluate a range of LLMs on classification, generation, and reasoning tasks. Our results reveal clear limitations of LLMs: models often confuse Drivelology with shallow nonsense, produce incoherent justifications, or miss implied rhetorical functions altogether. These findings highlight a deep representational gap in LLMs' pragmatic understanding and challenge the assumption that statistical fluency implies cognitive comprehension. We release our dataset and code to facilitate further research in modelling linguistic depth beyond surface-level coherence.

Large Language Models (LLMs) are increasingly used as proxies for human subjects in social science surveys, but their reliability and susceptibility to known human-like response biases, such as central tendency, opinion floating and primacy bias are poorly understood. This work investigates the response robustness of LLMs in normative survey contexts, we test nine LLMs on questions from the World Values Survey (WVS), applying a comprehensive set of ten perturbations to both question phrasing and answer option structure, resulting in over 167,000 simulated survey interviews. In doing so, we not only reveal LLMs' vulnerabilities to perturbations but also show that all tested models exhibit a consistent recency bias, disproportionately favoring the last-presented answer option. While larger models are generally more robust, all models remain sensitive to semantic variations like paraphrasing and to combined perturbations. This underscores the critical importance of prompt design and robustness testing when using LLMs to generate synthetic survey data.

- Large Language Models (LLMs) demonstrate impressive generative abilities across a wide range of tasks but continue to suffer from hallucinations --outputs that are fluent yet factually incorrect. This paper introduces a reference-free, token-level hallucination detection framework that identifies unreliable tokens by analyzing variance in log-probabilities across multiple stochastic generations. Unlike traditional methods that depend on external references or sentence-level verification, our approach is model-agnostic, interpretable, and computationally efficient, making it suitable for both real-time and post-hoc analysis. We evaluate the proposed method on three diverse datasets--SQuAD v2 (unanswerable questions), XSum (abstractive summarization), and TriviaQA (open-domain question answering)--using autoregressive models of increasing scale: GPT-Neo 125M, Falcon 1B, and Mistral 7B. Results show that token-level variance strongly correlates with hallucination behavior, revealing clear distinctions in uncertainty across model sizes. The framework maintains accuracy even under limited sampling conditions and introduces minimal computational overhead, supporting its practicality for lightweight deployment. Overall, this work provides a scalable, reproducible, and fine-grained diagnostic tool for detecting hallucinations in LLMs, with potential extensions to multilingual and real-time generation settings. Large language models (LLMs) have transformed natural language processing, powering tasks such as summarization, dialogue generation, and open-ended question answering.

Large Language Models demonstrate substantial promise for advancing scientific discovery, yet their deployment in disciplines demanding factual precision and specialized domain constraints presents significant challenges. Within molecular design for pharmaceutical development, these models can propose innovative molecular modifications but frequently generate chemically infeasible structures. We introduce VALID-Mol, a comprehensive framework that integrates chemical validation with LLM-driven molecular design, achieving an improvement in valid chemical structure generation from 3% to 83%. Our methodology synthesizes systematic prompt optimization, automated chemical verification, and domain-adapted fine-tuning to ensure dependable generation of synthesizable molecules with enhanced properties. Our contribution extends beyond implementation details to provide a transferable methodology for scientifically-constrained LLM applications with measurable reliability enhancements. Computational analyses indicate our framework generates promising synthesis candidates with up to 17-fold predicted improvements in target binding affinity while preserving synthetic feasibility.

Large language models (LLMs) have recently shown strong reasoning abilities in domains like mathematics, coding, and scientific problem-solving, yet their potential for ranking tasks, where prime examples include retrieval, recommender systems, and LLM routing, remains underexplored. Ranking requires complex reasoning across heterogeneous candidates, but existing LLM-based rankers are often domain-specific, tied to fixed backbones, and lack iterative refinement, limiting their ability to fully exploit LLMs' reasoning potential. To address these challenges, we propose R1-Ranker, a reasoning-incentive framework built on reinforcement learning, with two complementary designs: DRanker, which generates full rankings in one shot, and IRanker, which decomposes ranking into an iterative elimination process with step-wise rewards to encourage deeper reasoning. We evaluate unified R1-Rankers on nine datasets spanning recommendation, routing, and passage ranking, showing that IRanker-3B consistently achieves state-of-the-art performance, surpasses larger 7B models on some tasks, and yields a 15.7% average relative improvement. Ablation and generalization experiments further confirm the critical role of reinforcement learning and iterative reasoning, with IRanker-3B improving zero-shot performance by over 9% on out-of-domain tasks and reasoning traces boosting other LLMs by up to 22.87%. These results demonstrate that unifying diverse ranking tasks with a single reasoning-driven foundation model is both effective and essential for advancing LLM reasoning in ranking scenarios.

To ensure a balance between open access to justice and personal data protection, the South Korean judiciary mandates the de-identification of court judgments before they can be publicly disclosed. However, the current de-identification process is inadequate for handling court judgments at scale while adhering to strict legal requirements. Additionally, the legal definitions and categorizations of personal identifiers are vague and not well-suited for technical solutions. To tackle these challenges, we propose a de-identification framework called Thunder-DeID, which aligns with relevant laws and practices. Specifically, we (i) construct and release the first Korean legal dataset containing annotated judgments along with corresponding lists of entity mentions, (ii) introduce a systematic categorization of Personally Identifiable Information (PII), and (iii) develop an end-to-end deep neural network (DNN)-based de-identification pipeline. Our experimental results demonstrate that our model achieves state-of-the-art performance in the de-identification of court judgments.

Large Language Models (LLMs) have achieved remarkable progress, but their deployment has exposed critical vulnerabilities, particularly to jailbreak attacks that circumvent safety alignments. Guardrails--external defense mechanisms that monitor and control LLM interactions--have emerged as a promising solution. However, the current landscape of LLM guardrails is fragmented, lacking a unified taxonomy and comprehensive evaluation framework. In this Systematization of Knowledge (SoK) paper, we present the first holistic analysis of jailbreak guardrails for LLMs. We propose a novel, multi-dimensional taxonomy that categorizes guardrails along six key dimensions, and introduce a Security-Efficiency-Utility evaluation framework to assess their practical effectiveness. Through extensive analysis and experiments, we identify the strengths and limitations of existing guardrail approaches, provide insights into optimizing their defense mechanisms, and explore their universality across attack types. Our work offers a structured foundation for future research and development, aiming to guide the principled advancement and deployment of robust LLM guardrails. The code is available at https://github.com/xunguangwang/SoK4JailbreakGuardrails.

Large transformer-based language models dominate modern NLP, yet our understanding of how they encode linguistic information relies primarily on studies of early models like BERT and GPT-2. Building on classic BERTology work, we analyze 25 models spanning from classical architectures (BERT, DeBERTa, GPT-2) to modern large language models (Pythia, OLMo-2, Gemma-2, Qwen2.5, Llama-3.1), probing layer-by-layer representations across eight linguistic tasks in English. Consistent with earlier findings, we find that hierarchical organization persists in modern models: early layers capture syntax, middle layers handle semantics and entity-level information, and later layers encode discourse phenomena. We dive deeper, conducting an in-depth multilingual analysis of two specific linguistic properties - lexical identity and inflectional morphology - that help disentangle form from meaning. We find that lexical information concentrates linearly in early layers but becomes increasingly nonlinear deeper in the network, while inflectional information remains linearly accessible throughout all layers. Additional analyses of attention mechanisms, steering vectors, and pretraining checkpoints reveal where this information resides within layers, how it can be functionally manipulated, and how representations evolve during pretraining. Taken together, our findings suggest that, even with substantial advances in LLM technologies, transformer models learn to organize linguistic information in similar ways, regardless of model architecture, size, or training regime, indicating that these properties are important for next token prediction. Our code is available at https://github.com/ml5885/model_internal_sleuthing

Operations research (OR) is widely deployed to solve critical decision-making problems with complex objectives and constraints, impacting manufacturing, logistics, finance, and healthcare outcomes. While Large Language Models (LLMs) have shown promising results in various domains, their practical application in industry-relevant operations research (OR) problems presents significant challenges and opportunities. Preliminary industrial applications of LLMs for operations research face two critical deployment challenges: 1) Self-correction focuses on code syntax rather than mathematical accuracy, causing costly errors; 2) Complex expert selection creates unpredictable workflows that reduce transparency and increase maintenance costs, making them impractical for time-sensitive business applications. To address these business limitations, we introduce ORMind, a cognitive-inspired framework that enhances optimization through counterfactual reasoning. Our approach emulates human cognition, implementing an end-to-end workflow that systematically transforms requirements into mathematical models and executable solver code. It is currently being tested internally in Lenovo's AI Assistant, with plans to enhance optimization capabilities for both business and consumer customers. Experiments demonstrate that ORMind outperforms existing methods, achieving a 9.5\% improvement on the NL4Opt dataset and a 14.6\% improvement on the ComplexOR dataset.