llm - 2025_08
Navigation
Papers
Large language models implicitly encode preferences over human values, yet steering them often requires large training data. In this work, we investigate a simple approach: Can we reliably modify a model's value system in downstream behavior by training it to answer value survey questions accordingly? We first construct value profiles of several open-source LLMs by asking them to rate a series of value-related descriptions spanning 20 distinct human values, which we use as a baseline for subsequent experiments. We then investigate whether the value system of a model can be governed by fine-tuning on the value surveys. We evaluate the effect of finetuning on the model's behavior in two ways; first, we assess how answers change on in-domain, held-out survey questions. Second, we evaluate whether the model's behavior changes in out-of-domain settings (situational scenarios). To this end, we construct a contextualized moral judgment dataset based on Reddit posts and evaluate changes in the model's behavior in text-based adventure games. We demonstrate that our simple approach can not only change the model's answers to in-domain survey questions, but also produces substantial shifts (value alignment) in implicit downstream task behavior.
The increasing heterogeneity of student populations poses significant challenges for teachers, particularly in mathematics education, where cognitive, motivational, and emotional differences strongly influence learning outcomes. While AI-driven personalization tools have emerged, most remain performance-focused, offering limited support for teachers and neglecting broader pedagogical needs. This paper presents the FACET framework, a teacher-facing, large language model (LLM)-based multi-agent system designed to generate individualized classroom materials that integrate both cognitive and motivational dimensions of learner profiles. The framework comprises three specialized agents: (1) learner agents that simulate diverse profiles incorporating topic proficiency and intrinsic motivation, (2) a teacher agent that adapts instructional content according to didactical principles, and (3) an evaluator agent that provides automated quality assurance. We tested the system using authentic grade 8 mathematics curriculum content and evaluated its feasibility through a) automated agent-based assessment of output quality and b) exploratory feedback from K-12 in-service teachers. Results from ten internal evaluations highlighted high stability and alignment between generated materials and learner profiles, and teacher feedback particularly highlighted structure and suitability of tasks. The findings demonstrate the potential of multi-agent LLM architectures to provide scalable, context-aware personalization in heterogeneous classroom settings, and outline directions for extending the framework to richer learner profiles and real-world classroom trials.
LLM-based agents have emerged as transformative tools capable of executing complex tasks through iterative planning and action, achieving significant advancements in understanding and addressing user needs. Yet, their effectiveness remains limited in specialized domains such as mental health diagnosis, where they underperform compared to general applications. Current approaches to integrating diagnostic capabilities into LLMs rely on scarce, highly sensitive mental health datasets, which are challenging to acquire. These methods also fail to emulate clinicians' proactive inquiry skills, lack multi-turn conversational comprehension, and struggle to align outputs with expert clinical reasoning. To address these gaps, we propose DSM5AgentFlow, the first LLM-based agent workflow designed to autonomously generate DSM-5 Level-1 diagnostic questionnaires. By simulating therapist-client dialogues with specific client profiles, the framework delivers transparent, step-by-step disorder predictions, producing explainable and trustworthy results. This workflow serves as a complementary tool for mental health diagnosis, ensuring adherence to ethical and legal standards. Through comprehensive experiments, we evaluate leading LLMs across three critical dimensions: conversational realism, diagnostic accuracy, and explainability. Our datasets and implementations are fully open-sourced.
The Roots of International Perceptions: Simulating US Attitude Changes Towards China with LLM Agents
The rise of LLMs poses new possibilities in modeling opinion evolution, a long-standing task in simulation, by leveraging advanced reasoning abilities to recreate complex, large-scale human cognitive trends. While most prior works focus on opinion evolution surrounding specific isolated events or the views within a country, ours is the first to model the large-scale attitude evolution of a population representing an entire country towards another -- US citizens' perspectives towards China. To tackle the challenges of this broad scenario, we propose a framework that integrates media data collection, user profile creation, and cognitive architecture for opinion updates to successfully reproduce the real trend of US attitudes towards China over a 20-year period from 2005 to today. We also leverage LLMs' capabilities to introduce debiased media exposure, extracting neutral events from typically subjective news contents, to uncover the roots of polarized opinion formation, as well as a devils advocate agent to help explain the rare reversal from negative to positive attitudes towards China, corresponding with changes in the way Americans obtain information about the country. The simulation results, beyond validating our framework architecture, also reveal the impact of biased framing and selection bias in shaping attitudes. Overall, our work contributes to a new paradigm for LLM-based modeling of cognitive behaviors in a large-scale, long-term, cross-border social context, providing insights into the formation of international biases and offering valuable implications for media consumers to better understand the factors shaping their perspectives, and ultimately contributing to the larger social need for bias reduction and cross-cultural tolerance.
Large Language Models (LLMs) are highly sensitive to subtle, non-semantic variations in prompt phrasing and formatting. In this work, we present the first systematic evaluation of 5 methods for improving prompt robustness within a unified experimental framework. We benchmark these techniques on 8 models from Llama, Qwen and Gemma families across 52 tasks from Natural Instructions dataset. Our evaluation covers robustness methods from both fine-tuned and in-context learning paradigms, and tests their generalization against multiple types of distribution shifts. Finally, we extend our analysis to GPT-4.1 and DeepSeek V3 to assess frontier models' current robustness to format perturbations. Our findings offer actionable insights into the relative effectiveness of these robustness methods, enabling practitioners to make informed decisions when aiming for stable and reliable LLM performance in real-world applications. Code: https://github.com/AIRI-Institute/when-punctuation-matters.
Formal logic enables computers to reason in natural language by representing sentences in symbolic forms and applying rules to derive conclusions. However, in what our study characterizes as "rulebreaker" scenarios, this method can lead to conclusions that are typically not inferred or accepted by humans given their common sense and factual knowledge. Inspired by works in cognitive science, we create RULEBREAKERS, the first dataset for rigorously evaluating the ability of large language models (LLMs) to recognize and respond to rulebreakers (versus non-rulebreakers) in a human-like manner. Evaluating seven LLMs, we find that most models, including GPT-4o, achieve mediocre accuracy on RULEBREAKERS and exhibit some tendency to over-rigidly apply logical rules unlike what is expected from typical human reasoners. Further analysis suggests that this apparent failure is potentially associated with the models' poor utilization of their world knowledge and their attention distribution patterns. Whilst revealing a limitation of current LLMs, our study also provides a timely counterbalance to a growing body of recent works that propose methods relying on formal logic to improve LLMs' general reasoning capabilities, highlighting their risk of further increasing divergence between LLMs and human-like reasoning.
The proliferation of high-quality text from Large Language Models (LLMs) demands reliable and efficient detection methods. While existing training-free approaches show promise, they often rely on surface-level statistics and overlook fundamental signal properties of the text generation process. In this work, we reframe detection as a signal processing problem, introducing a novel paradigm that analyzes the sequence of token log-probabilities in the frequency domain. By systematically analyzing the signal's spectral properties using the global Discrete Fourier Transform (DFT) and the local Short-Time Fourier Transform (STFT), we find that human-written text consistently exhibits significantly higher spectral energy. This higher energy reflects the larger-amplitude fluctuations inherent in human writing compared to the suppressed dynamics of LLM-generated text. Based on this key insight, we construct SpecDetect, a detector built on a single, robust feature from the global DFT: DFT total energy. We also propose an enhanced version, SpecDetect++, which incorporates a sampling discrepancy mechanism to further boost robustness. Extensive experiments demonstrate that our approach outperforms the state-of-the-art model while running in nearly half the time. Our work introduces a new, efficient, and interpretable pathway for LLM-generated text detection, showing that classical signal processing techniques offer a surprisingly powerful solution to this modern challenge.
Quantization is an essential and popular technique for improving the accessibility of large language models (LLMs) by reducing memory usage and computational costs while maintaining performance. In this study, we apply 4-bit Group Scaling Quantization (GSQ) and Generative Pretrained Transformer Quantization (GPTQ) to LLaMA 1B, Qwen 0.5B, and PHI 1.5B, evaluating their impact across multiple NLP tasks. We benchmark these models on MS MARCO (Information Retrieval), BoolQ (Boolean Question Answering), and GSM8K (Mathematical Reasoning) datasets, assessing both accuracy and efficiency across various tasks. The study measures the trade-offs between model compression and task performance, analyzing key evaluation metrics, namely accuracy, inference latency, and throughput (total output tokens generated per second), providing insights into the suitability of low-bit quantization for real-world deployment. Using the results, users can then make suitable decisions based on the specifications that need to be met. We discuss the pros and cons of GSQ and GPTQ techniques on models of different sizes, which also serve as a benchmark for future experiments.
Logging code is written by developers to capture system runtime behavior and plays a vital role in debugging, performance analysis, and system monitoring. However, defects in logging code can undermine the usefulness of logs and lead to misinterpretations. Although prior work has identified several logging defect patterns and provided valuable insights into logging practices, these studies often focus on a narrow range of defect patterns derived from limited sources (e.g., commit histories) and lack a systematic and comprehensive analysis. Moreover, large language models (LLMs) have demonstrated promising generalization and reasoning capabilities across a variety of code-related tasks, yet their potential for detecting logging code defects remains largely unexplored. In this paper, we derive a comprehensive taxonomy of logging code defects, which encompasses seven logging code defect patterns with 14 detailed scenarios. We further construct a benchmark dataset, \dataset, consisting of 164 developer-verified real-world logging defects. Then we propose an automated framework that leverages various prompting strategies and contextual information to evaluate LLMs' capability in detecting and reasoning logging code defects. Experimental results reveal that LLMs generally struggle to accurately detect and reason logging code defects based on the source code only. However, incorporating proper knowledge (e.g., detailed scenarios of defect patterns) can lead to 10.9\% improvement in detection accuracy. Overall, our findings provide actionable guidance for practitioners to avoid common defect patterns and establish a foundation for improving LLM-based reasoning in logging code defect detection.
Existing work on large language model (LLM) personalization assigned different responding roles to LLMs, but overlooked the diversity of queriers. In this work, we propose a new form of querier-aware LLM personalization, generating different responses even for the same query from different queriers. We design a dual-tower model architecture with a cross-querier general encoder and a querier-specific encoder. We further apply contrastive learning with multi-view augmentation, pulling close the dialogue representations of the same querier, while pulling apart those of different queriers. To mitigate the impact of query diversity on querier-contrastive learning, we cluster the dialogues based on query similarity and restrict the scope of contrastive learning within each cluster. To address the lack of datasets designed for querier-aware personalization, we also build a multi-querier dataset from English and Chinese scripts, as well as WeChat records, called MQDialog, containing 173 queriers and 12 responders. Extensive evaluations demonstrate that our design significantly improves the quality of personalized response generation, achieving relative improvement of 8.4% to 48.7% in ROUGE-L scores and winning rates ranging from 54% to 82% compared with various baseline methods.
The integration of wireless communications and Large Language Models (LLMs) is poised to unlock ubiquitous intelligent services, yet deploying them in wireless edge-device collaborative environments presents a critical trade-off between inference quality and end-to-end latency. A fundamental mismatch exists between task complexity and resource allocation: offloading simple queries invites prohibitive latency, while on-device models lack the capacity for demanding computations. To address this challenge, we propose a dynamic, quality-latency aware routing framework that orchestrates inference between a lightweight model on the mobile device and a powerful model on the edge server. Our framework employs two distinct cost models: for single-turn queries, it fuses a BERT-predicted semantic score with communication and computation overheads; for multi-turn dialogues, it further quantifies context-aware costs arising from model switching and KV-cache management. While maintaining full inference quality, extensive experiments demonstrate that our framework cuts average response latency by 5-15% and reduces large model invocations by 10-20% against competitive baselines on MMLU, GSM8K, and MT-Bench-101 benchmarks.
LLMs increasingly exhibit over-refusal behavior, where safety mechanisms cause models to reject benign instructions that superficially resemble harmful content. This phenomena diminishes utility in production applications that repeatedly rely on common prompt templates or applications that frequently rely on LLMs for specific tasks (e.g. sentiment analysis, language translation). Through comprehensive evaluation, we demonstrate that LLMs still tend to refuse responses to harmful instructions when those instructions are reframed to appear as benign tasks. Our mechanistic analysis reveal that LLMs follow distinct "constellation" patterns in embedding space as representations traverse layers, with each task maintaining consistent trajectories that shift predictably between refusal and non-refusal cases. We introduce SafeConstellations, an inference-time trajectory-shifting approach that tracks task-specific trajectory patterns and guides representations toward non-refusal pathways. By selectively guiding model behavior only on tasks prone to over-refusal, and by preserving general model behavior, our method reduces over-refusal rates by up to 73% with minimal impact on utility-offering a principled approach to mitigating over-refusals.
With the significant success achieved by large language models (LLMs) like LLaMA, edge computing-based LLM inference services for mobile and PC are in high demand for data privacy. However, different edge platforms have different hardware characteristics and the large demand for memory capacity and bandwidth makes it very challenging to deploy and benchmark LLMs on edge devices. In this paper, we introduce a benchmarking tool named ELIB (edge LLM inference benchmarking) to evaluate LLM inference performance of different edge platforms, and propose a novel metric named MBU to indicate the percentage of the theoretically efficient use of available memory bandwidth for a specific model running on edge hardware to optimize memory usage. We deploy ELIB on three edge platforms and benchmark using five quantized models to optimize MBU in combination with other metrics such as FLOPS, throughput, latency and accuracy. And we analyze the results to derive the key factors, constraints, unpredictability in optimizing MBU that can guide deploying LLMs on more edge platforms.
Large language models (LLMs) have demonstrated potential in reasoning tasks, but their performance on linguistics puzzles remains consistently poor. These puzzles, often derived from Linguistics Olympiad (LO) contests, provide a minimal contamination environment to assess LLMs' linguistic reasoning abilities across low-resource languages. This work analyses LLMs' performance on 629 problems across 41 low-resource languages by labelling each with linguistically informed features to unveil weaknesses. Our analyses show that LLMs struggle with puzzles involving higher morphological complexity and perform better on puzzles involving linguistic features that are also found in English. We also show that splitting words into morphemes as a pre-processing step improves solvability, indicating a need for more informed and language-specific tokenisers. These findings thus offer insights into some challenges in linguistic reasoning and modelling of low-resource languages.
Instruction fine-tuned large language models (LLMs) enable a simple zero-shot or few-shot prompting paradigm, also known as in-context learning, for building prediction models. This convenience, combined with continued advances in LLM capability, has the potential to drive their adoption across a broad range of domains, including high-stakes applications where group fairness -- preventing disparate impacts across demographic groups -- is essential. The majority of existing approaches to enforcing group fairness on LLM-based classifiers rely on traditional fair algorithms applied via model fine-tuning or head-tuning on final-layer embeddings, but they are no longer applicable to closed-weight LLMs under the in-context learning setting, which include some of the most capable commercial models today, such as GPT-4, Gemini, and Claude. In this paper, we propose a framework for deriving fair classifiers from closed-weight LLMs via prompting: the LLM is treated as a feature extractor, and features are elicited from its probabilistic predictions (e.g., token log probabilities) using prompts strategically designed for the specified fairness criterion to obtain sufficient statistics for fair classification; a fair algorithm is then applied to these features to train a lightweight fair classifier in a post-hoc manner. Experiments on five datasets, including three tabular ones, demonstrate strong accuracy-fairness tradeoffs for the classifiers derived by our framework from both open-weight and closed-weight LLMs; in particular, our framework is data-efficient and outperforms fair classifiers trained on LLM embeddings (i.e., head-tuning) or from scratch on raw tabular features.
Large Language Models (LLMs) have shown significant potential in automating code generation tasks offering new opportunities across software engineering domains. However, their practical application remains limited due to hallucinations - outputs that appear plausible but are factually incorrect, unverifiable or nonsensical. This paper investigates hallucination phenomena in the context of code generation with a specific focus on the automotive domain. A case study is presented that evaluates multiple code LLMs for three different prompting complexities ranging from a minimal one-liner prompt to a prompt with Covesa Vehicle Signal Specifications (VSS) as additional context and finally to a prompt with an additional code skeleton. The evaluation reveals a high frequency of syntax violations, invalid reference errors and API knowledge conflicts in state-of-the-art models GPT-4.1, Codex and GPT-4o. Among the evaluated models, only GPT-4.1 and GPT-4o were able to produce a correct solution when given the most context-rich prompt. Simpler prompting strategies failed to yield a working result, even after multiple refinement iterations. These findings highlight the need for effective mitigation techniques to ensure the safe and reliable use of LLM generated code, especially in safety-critical domains such as automotive software systems.
There is a growing interest in utilizing large-scale language models (LLMs) to advance next-generation Recommender Systems (RecSys), driven by their outstanding language understanding and in-context learning capabilities. In this scenario, tokenizing (i.e., indexing) users and items becomes essential for ensuring a seamless alignment of LLMs with recommendations. While several studies have made progress in representing users and items through textual contents or latent representations, challenges remain in efficiently capturing high-order collaborative knowledge into discrete tokens that are compatible with LLMs. Additionally, the majority of existing tokenization approaches often face difficulties in generalizing effectively to new/unseen users or items that were not in the training corpus. To address these challenges, we propose a novel framework called TokenRec, which introduces not only an effective ID tokenization strategy but also an efficient retrieval paradigm for LLM-based recommendations. Specifically, our tokenization strategy, Masked Vector-Quantized (MQ) Tokenizer, involves quantizing the masked user/item representations learned from collaborative filtering into discrete tokens, thus achieving a smooth incorporation of high-order collaborative knowledge and a generalizable tokenization of users and items for LLM-based RecSys. Meanwhile, our generative retrieval paradigm is designed to efficiently recommend top-$K$ items for users to eliminate the need for the time-consuming auto-regressive decoding and beam search processes used by LLMs, thus significantly reducing inference time. Comprehensive experiments validate the effectiveness of the proposed methods, demonstrating that TokenRec outperforms competitive benchmarks, including both traditional recommender systems and emerging LLM-based recommender systems.
Large Language Models (LLMs) increasingly exhibit over-refusal - erroneously rejecting benign queries due to overly conservative safety measures - a critical functional flaw that undermines their reliability and usability. Current methods for testing this behavior are demonstrably inadequate, suffering from flawed benchmarks and limited test generation capabilities, as highlighted by our empirical user study. To the best of our knowledge, this paper introduces the first evolutionary testing framework, ORFuzz, for the systematic detection and analysis of LLM over-refusals. ORFuzz uniquely integrates three core components: (1) safety category-aware seed selection for comprehensive test coverage, (2) adaptive mutator optimization using reasoning LLMs to generate effective test cases, and (3) OR-Judge, a human-aligned judge model validated to accurately reflect user perception of toxicity and refusal. Our extensive evaluations demonstrate that ORFuzz generates diverse, validated over-refusal instances at a rate (6.98% average) more than double that of leading baselines, effectively uncovering vulnerabilities. Furthermore, ORFuzz's outputs form the basis of ORFuzzSet, a new benchmark of 1,855 highly transferable test cases that achieves a superior 63.56% average over-refusal rate across 10 diverse LLMs, significantly outperforming existing datasets. ORFuzz and ORFuzzSet provide a robust automated testing framework and a valuable community resource, paving the way for developing more reliable and trustworthy LLM-based software systems.
Multimodal large language models (MLLMs) extend LLMs to handle images, videos, and audio by incorporating feature extractors and projection modules. However, these additional components -- combined with complex inference pipelines and heterogeneous workloads -- introduce significant inference overhead. Therefore, efficiently serving MLLMs remains a major challenge. Current tightly coupled serving architectures struggle to distinguish between mixed request types or adapt parallelism strategies to different inference stages, leading to increased time-to-first-token (TTFT) latency and poor resource utilization. To address this, we introduce Elastic Multimodal Parallelism (EMP), a new serving paradigm that elastically adapts to resource heterogeneity across request types and inference stages. Building upon EMP, we develop ElasticMM, an MLLM serving system that (1) separates requests into independent modality groups with dynamic resource allocation via a modality-aware load balancer; (2) decouples inference stages and enables parallelism adjustment and adaptive scaling via elastic partition scheduling; and (3) improves inference efficiency through unified multimodal prefix caching and non-blocking encoding. Experiments on diverse real-world datasets show that ElasticMM outperforms state-of-the-art (SOTA) serving systems, reducing TTFT by up to 4.2x and achieving 3.2-4.5x higher throughput while meeting service-level objectives (SLOs).
This survey provides a comprehensive review of research on multi-turn dialogue systems, with a particular focus on multi-turn dialogue systems based on large language models (LLMs). This paper aims to (a) give a summary of existing LLMs and approaches for adapting LLMs to downstream tasks; (b) elaborate recent advances in multi-turn dialogue systems, covering both LLM-based open-domain dialogue (ODD) and task-oriented dialogue (TOD) systems, along with datasets and evaluation metrics; (c) discuss some future emphasis and recent research problems arising from the development of LLMs and the increasing demands on multi-turn dialogue systems.
Recent advances in video-based multimodal large language models (Video-LLMs) have significantly improved video understanding by processing videos as sequences of image frames. However, many existing methods treat frames independently in the vision backbone, lacking explicit temporal modeling, which limits their ability to capture dynamic patterns and efficiently handle long videos. To address these limitations, we introduce STORM (Spatiotemporal TOken Reduction for Multimodal LLMs), a novel architecture incorporating a dedicated temporal encoder between the image encoder and the LLM. Our temporal encoder leverages the Mamba State Space Model to integrate temporal information into image tokens, generating enriched representations that preserve inter-frame dynamics across the entire video sequence. This enriched encoding not only enhances video reasoning capabilities but also enables effective token reduction strategies, including test-time sampling and training-based temporal and spatial pooling, substantially reducing computational demands on the LLM without sacrificing key temporal information. By integrating these techniques, our approach simultaneously reduces training and inference latency while improving performance, enabling efficient and robust video understanding over extended temporal contexts. Extensive evaluations show that STORM achieves state-of-the-art results across various long video understanding benchmarks (more than 5% improvement on MLVU and LongVideoBench) while reducing the computation costs by up to $8\times$ and the decoding latency by 2.4-2.9$\times$ for the fixed numbers of input frames. Project page is available at https://research.nvidia.com/labs/lpr/storm
The rise of long-context Large Language Models (LLMs) amplifies memory and bandwidth demands during autoregressive decoding, as the Key-Value (KV) cache grows with each generated token. Low-bit KV-cache quantization (e.g., 4-bit or 2-bit) can reduce memory footprint while preserving accuracy, but existing systems suffer from slow decoding due to their exclusive reliance on CUDA cores, neglecting Tensor Cores (the primary source of compute on modern GPUs). We present BitDecoding, a new long-context LLM inference system with a low-bit KV cache. BitDecoding enables efficient low-bit KV-cache decoding by cooperatively leveraging CUDA cores and Tensor Cores. It introduces methods for automatically inducing optimized layouts to exploit Tensor Cores, along with warp-level parallelization strategies for dequantization. For unified system support, BitDecoding includes a query transformation module supporting diverse attention variants, a quantization kernel that supports both tensor-wise and channel-wise scaling used in various quantization algorithms with high performance, and a dequantization kernel with a software-defined pipeline to coordinate CUDA and Tensor Cores execution for mixed-precision operations. Evaluated on RTX 4090, A100, and H100, BitDecoding accelerates decoding by up to 7.5x, 4.8x, and 8.9x, respectively, over FP16 FlashDecoding-v2, and surpasses the state-of-the-art low-bit system QServe by up to 4.3x. On LLaMA-3.1-8B with a 128K context, BitDecoding reduces single-batch decoding latency by 3x, showing substantial improvements for long-context generation. The code is available at https://github.com/DD-DuDa/BitDecoding.
Despite large language models (LLMs) have achieved remarkable success, their prefix-only prompting paradigm and sequential generation process offer limited flexibility for bidirectional information. Diffusion large language models (dLLMs) present new opportunities through their bidirectional attention mechanisms and iterative refinement processes, enabling more flexible in-place prompting strategies. We introduce ICE (In-Place Chain-of-Thought Prompting with Early Exit), a novel framework that transforms prefix-only prompting into in-place prompting specifically designed for dLLMs. ICE integrates in-place prompts directly within masked token positions during iterative refinement and employs a confidence-aware early exit mechanism to significantly reduce computational overhead. Extensive experiments demonstrate ICE's effectiveness, achieving up to 17.29% accuracy improvement with 4.12$\times$ speedup on GSM8K, and up to 276.67$\times$ acceleration on MMLU while maintaining competitive performance.
Recent advancements in post-training Large Language Models (LLMs), particularly through Reinforcement Learning (RL) and preference optimization methods, are key drivers for enhancing their reasoning capabilities. However, these methods are often plagued by low sample efficiency and a susceptibility to primacy bias, where overfitting to initial experiences degrades policy quality and damages the learning process. To address these challenges, we introduce LLM optimization with Reset Replay (LoRR), a general and powerful plugin designed to enhance sample efficiency in any preference-based optimization framework. LoRR core mechanism enables training at a high replay number, maximizing the utility of each collected data batch. To counteract the risk of overfitting inherent in high-replay training, LoRR incorporates a periodic reset strategy with reusing initial data, which preserves network plasticity. Furthermore, it leverages a hybrid optimization objective, combining supervised fine-tuning (SFT) and preference-based losses to further bolster data exploitation. Our extensive experiments demonstrate that LoRR significantly boosts the performance of various preference optimization methods on both mathematical and general reasoning benchmarks. Notably, an iterative DPO approach augmented with LoRR achieves comparable performance on challenging math tasks, outperforming some complex and computationally intensive RL-based algorithms. These findings highlight that LoRR offers a practical, sample-efficient, and highly effective paradigm for LLM finetuning, unlocking greater performance from limited data.
Effective incident response (IR) is critical for mitigating cyber threats, yet security teams are overwhelmed by alert fatigue, high false-positive rates, and the vast volume of unstructured Cyber Threat Intelligence (CTI) documents. While CTI holds immense potential for enriching security operations, its extensive and fragmented nature makes manual analysis time-consuming and resource-intensive. To bridge this gap, we introduce a novel Retrieval-Augmented Generation (RAG)-based framework that leverages Large Language Models (LLMs) to automate and enhance IR by integrating dynamically retrieved CTI. Our approach introduces a hybrid retrieval mechanism that combines NLP-based similarity searches within a CTI vector database with standardized queries to external CTI platforms, facilitating context-aware enrichment of security alerts. The augmented intelligence is then leveraged by an LLM-powered response generation module, which formulates precise, actionable, and contextually relevant incident mitigation strategies. We propose a dual evaluation paradigm, wherein automated assessment using an auxiliary LLM is systematically cross-validated by cybersecurity experts. Empirical validation on real-world and simulated alerts demonstrates that our approach enhances the accuracy, contextualization, and efficiency of IR, alleviating analyst workload and reducing response latency. This work underscores the potential of LLM-driven CTI fusion in advancing autonomous security operations and establishing a foundation for intelligent, adaptive cybersecurity frameworks.
Estimating treatment effects (TE) from observational data is a critical yet complex task in many fields, from healthcare and economics to public policy. While recent advances in machine learning and causal inference have produced powerful estimation techniques, their adoption remains limited due to the need for deep expertise in causal assumptions, adjustment strategies, and model selection. In this paper, we introduce CATE-B, an open-source co-pilot system that uses large language models (LLMs) within an agentic framework to guide users through the end-to-end process of treatment effect estimation. CATE-B assists in (i) constructing a structural causal model via causal discovery and LLM-based edge orientation, (ii) identifying robust adjustment sets through a novel Minimal Uncertainty Adjustment Set criterion, and (iii) selecting appropriate regression methods tailored to the causal structure and dataset characteristics. To encourage reproducibility and evaluation, we release a suite of benchmark tasks spanning diverse domains and causal complexities. By combining causal inference with intelligent, interactive assistance, CATE-B lowers the barrier to rigorous causal analysis and lays the foundation for a new class of benchmarks in automated treatment effect estimation.
This paper presents a feasibility study on the deployment of a European Deep Inference Fabric (eDIF), an NDIF-compatible infrastructure designed to support mechanistic interpretability research on large language models. The need for widespread accessibility of LLM interpretability infrastructure in Europe drives this initiative to democratize advanced model analysis capabilities for the research community. The project introduces a GPU-based cluster hosted at Ansbach University of Applied Sciences and interconnected with partner institutions, enabling remote model inspection via the NNsight API. A structured pilot study involving 16 researchers from across Europe evaluated the platform's technical performance, usability, and scientific utility. Users conducted interventions such as activation patching, causal tracing, and representation analysis on models including GPT-2 and DeepSeek-R1-70B. The study revealed a gradual increase in user engagement, stable platform performance throughout, and a positive reception of the remote experimentation capabilities. It also marked the starting point for building a user community around the platform. Identified limitations such as prolonged download durations for activation data as well as intermittent execution interruptions are addressed in the roadmap for future development. This initiative marks a significant step towards widespread accessibility of LLM interpretability infrastructure in Europe and lays the groundwork for broader deployment, expanded tooling, and sustained community collaboration in mechanistic interpretability research.
Solidity, the dominant smart contract language for Ethereum, has rapidly evolved with frequent version updates to enhance security, functionality, and developer experience. However, these continual changes introduce significant challenges, particularly in compilation errors, code migration, and maintenance. Therefore, we conduct an empirical study to investigate the challenges in the Solidity version evolution and reveal that 81.68% of examined contracts encounter errors when compiled across different versions, with 86.92% of compilation errors. To mitigate these challenges, we conducted a systematic evaluation of large language models (LLMs) for resolving Solidity compilation errors during version migrations. Our empirical analysis across both open-source (LLaMA3, DeepSeek) and closed-source (GPT-4o, GPT-3.5-turbo) LLMs reveals that although these models exhibit error repair capabilities, their effectiveness diminishes significantly for semantic-level issues and shows strong dependency on prompt engineering strategies. This underscores the critical need for domain-specific adaptation in developing reliable LLM-based repair systems for smart contracts. Building upon these insights, we introduce SMCFIXER, a novel framework that systematically integrates expert knowledge retrieval with LLM-based repair mechanisms for Solidity compilation error resolution. The architecture comprises three core phases: (1) context-aware code slicing that extracts relevant error information; (2) expert knowledge retrieval from official documentation; and (3) iterative patch generation for Solidity migration. Experimental validation across Solidity version migrations demonstrates our approach's statistically significant 24.24% improvement over baseline GPT-4o on real-world datasets, achieving near-perfect 96.97% accuracy.
Contemporary spatial services such as online maps predominantly rely on user queries for location searches. However, the user experience is limited when performing complex tasks, such as searching for a group of locations simultaneously. In this study, we examine the extended scenario known as Spatial Exemplar Query (SEQ), where multiple relevant locations are jointly searched based on user-specified examples. We introduce SEQ-GPT, a spatial query system powered by Large Language Models (LLMs) towards more versatile SEQ search using natural language. The language capabilities of LLMs enable unique interactive operations in the SEQ process, including asking users to clarify query details and dynamically adjusting the search based on user feedback. We also propose a tailored LLM adaptation pipeline that aligns natural language with structured spatial data and queries through dialogue synthesis and multi-model cooperation. SEQ-GPT offers an end-to-end demonstration for broadening spatial search with realistic data and application scenarios.
Question answering over Scholarly Knowledge Graphs (SKGs) remains a challenging task due to the complexity of scholarly content and the intricate structure of these graphs. Large Language Model (LLM) approaches could be used to translate natural language questions (NLQs) into SPARQL queries; however, these LLM-based approaches struggle with SPARQL query generation due to limited exposure to SKG-specific content and the underlying schema. We identified two main types of errors in the LLM-generated SPARQL queries: (i) structural inconsistencies, such as missing or redundant triples in the queries, and (ii) semantic inaccuracies, where incorrect entities or properties are shown in the queries despite a correct query structure. To address these issues, we propose FIRESPARQL, a modular framework that supports fine-tuned LLMs as a core component, with optional context provided via retrieval-augmented generation (RAG) and a SPARQL query correction layer. We evaluate the framework on the SciQA Benchmark using various configurations (zero-shot, zero-shot with RAG, one-shot, fine-tuning, and fine-tuning with RAG) and compare the performance with baseline and state-of-the-art approaches. We measure query accuracy using BLEU and ROUGE metrics, and query result accuracy using relaxed exact match(RelaxedEM), with respect to the gold standards containing the NLQs, SPARQL queries, and the results of the queries. Experimental results demonstrate that fine-tuning achieves the highest overall performance, reaching 0.90 ROUGE-L for query accuracy and 0.85 RelaxedEM for result accuracy on the test set.
The increasing scale and complexity of large language models (LLMs) pose significant inference latency challenges, primarily due to their autoregressive decoding paradigm characterized by the sequential nature of next-token prediction. By re-examining the outputs of autoregressive models, we observed that some segments exhibit parallelizable structures, which we term intrinsic parallelism. Decoding each parallelizable branch simultaneously (i.e. parallel decoding) can significantly improve the overall inference speed of LLMs. In this paper, we propose an Adaptive Serial-Parallel Decoding (ASPD), which addresses two core challenges: automated construction of parallelizable data and efficient parallel decoding mechanism. More specifically, we introduce a non-invasive pipeline that automatically extracts and validates parallelizable structures from the responses of autoregressive models. To empower efficient adaptive serial-parallel decoding, we implement a Hybrid Decoding Engine which enables seamless transitions between serial and parallel decoding modes while maintaining a reusable KV cache, maximizing computational efficiency. Extensive evaluations across General Tasks, Retrieval-Augmented Generation, Mathematical Reasoning, demonstrate that ASPD achieves unprecedented performance in both effectiveness and efficiency. Notably, on Vicuna Bench, our method achieves up to 3.19x speedup (1.85x on average) while maintaining response quality within 1% difference compared to autoregressive models, realizing significant acceleration without compromising generation quality. Our framework sets a groundbreaking benchmark for efficient LLM parallel inference, paving the way for its deployment in latency-sensitive applications such as AI-powered customer service bots and answer retrieval engines.
Open-source software (OSS) has experienced a surge in popularity, attributed to its collaborative development model and cost-effective nature. However, the adoption of specific software versions in development projects may introduce security risks when these versions bring along vulnerabilities. Current methods of identifying vulnerable versions typically analyze and extract the code features involved in vulnerability patches using static analysis with pre-defined rules. They then use code clone detection to identify the vulnerable versions. These methods are hindered by imprecision due to (1) the exclusion of vulnerability-irrelevant code in the analysis and (2) the inadequacy of code clone detection. This paper presents VERCATION, an approach designed to identify vulnerable versions of OSS written in C/C++. VERCATION combines program slicing with a Large Language Model (LLM) to identify vulnerability-relevant code from vulnerability patches. It then backtracks historical commits to gather previous modifications of identified vulnerability-relevant code. We propose code clone detection based on expanded and normalized ASTs to compare the differences between pre-modification and post-modification code, thereby locating the vulnerability-introducing commit (vic) and enabling the identification of the vulnerable versions between the vulnerability-fixing commit and the vic. We curate a dataset linking 122 OSS vulnerabilities and 1,211 versions to evaluate VERCATION. On this dataset, our approach achieves an F1 score of 93.1%, outperforming current state-of-the-art methods. More importantly, VERCATION detected 202 incorrect vulnerable OSS versions in NVD reports.
Evaluating large language models (LLMs) in complex decision-making is essential for advancing AI's ability for strategic planning and real-time adaptation. However, existing benchmarks for tasks like StarCraft II fail to capture the game's full complexity, such as its complete game context, diverse action spaces, and all playable races. To address this gap, we present SC2Arena, a benchmark that fully supports all playable races, low-level action spaces, and optimizes text-based observations to tackle spatial reasoning challenges. Complementing this, we introduce StarEvolve, a hierarchical framework that integrates strategic planning with tactical execution, featuring iterative self-correction and continuous improvement via fine-tuning on high-quality gameplay data. Its key components include a Planner-Executor-Verifier structure to break down gameplay, and a scoring system for selecting high-quality training samples. Comprehensive analysis using SC2Arena provides valuable insights into developing generalist agents that were not possible with previous benchmarks. Experimental results also demonstrate that our proposed StarEvolve achieves superior performance in strategic planning. Our code, environment, and algorithms are publicly available.
Large Language Models (LLMs) have demonstrated remarkable capabilities in natural language processing tasks, yet their application to graph structure analysis, particularly in community search, remains underexplored. Community search, a fundamental task in graph analysis, aims to identify groups of nodes with dense interconnections, which is crucial for understanding the macroscopic structure of graphs. In this paper, we propose GraphCS, a comprehensive benchmark designed to evaluate the performance of LLMs in community search tasks. Our experiments reveal that while LLMs exhibit preliminary potential, they frequently fail to return meaningful results and suffer from output bias. To address these limitations, we introduce CS-Agent, a dual-agent collaborative framework to enhance LLM-based community search. CS-Agent leverages the complementary strengths of two LLMs acting as Solver and Validator. Through iterative feedback and refinement, CS-Agent dynamically refines initial results without fine-tuning or additional training. After the multi-round dialogue, Decider module selects the optimal community. Extensive experiments demonstrate that CS-Agent significantly improves the quality and stability of identified communities compared to baseline methods. To our knowledge, this is the first work to apply LLMs to community search, bridging the gap between LLMs and graph analysis while providing a robust and adaptive solution for real-world applications.
Large language models (LLMs) have achieved significant performance gains via scaling up model sizes and/or data. However, recent evidence suggests diminishing returns from such approaches, motivating scaling the computation spent at inference time. Existing inference-time scaling methods, usually with reward models, cast the task as a search problem, which tends to be vulnerable to reward hacking as a consequence of approximation errors in reward models. In this paper, we instead cast inference-time scaling as a probabilistic inference task and leverage sampling-based techniques to explore the typical set of the state distribution of a state-space model with an approximate likelihood, rather than optimize for its mode directly. We propose a novel inference-time scaling approach by adapting particle-based Monte Carlo methods to this task. Our empirical evaluation demonstrates that our methods have a 4-16x better scaling rate over our deterministic search counterparts on various challenging mathematical reasoning tasks. Using our approach, we show that Qwen2.5-Math-1.5B-Instruct can surpass GPT-4o accuracy in only 4 rollouts, while Qwen2.5-Math-7B-Instruct scales to o1 level accuracy in only 32 rollouts. Our work not only presents an effective method to inference-time scaling, but also connects the rich literature in probabilistic inference with inference-time scaling of LLMs to develop more robust algorithms in future work. Code, videos, and further information available at https://probabilistic-inference-scaling.github.io.
Although LLM inference has emerged as a critical workload for many downstream applications, efficiently inferring LLMs is challenging due to the substantial memory footprint and bandwidth requirements. In parallel, compute capabilities have steadily outpaced both memory capacity and bandwidth over the last few decades, a trend that remains evident in modern GPU hardware and exacerbates the challenge of LLM inference. As such, new algorithms are emerging that trade increased computation for reduced memory operations. To that end, we present XQuant, which takes advantage of this trend, enabling an order-of-magnitude reduction in memory consumption through low-bit quantization with substantial accuracy benefits relative to state-of-the-art KV cache quantization methods. We accomplish this by quantizing and caching the layer input activations X, instead of using standard KV caching, and then rematerializing the Keys and Values on-the-fly during inference. This results in an immediate 2$\times$ memory savings compared to KV caching. By applying XQuant, we achieve up to $\sim 7.7\times$ memory savings with $<0.1$ perplexity degradation compared to the FP16 baseline. Furthermore, our approach leverages the fact that X values are similar across layers. Building on this observation, we introduce XQuant-CL, which exploits the cross-layer similarity in the X embeddings for extreme compression. Across different models, XQuant-CL attains up to 10$\times$ memory savings relative to the FP16 baseline with only 0.01 perplexity degradation, and 12.5$\times$ memory savings with only $0.1$ perplexity degradation. XQuant exploits the rapidly increasing compute capabilities of hardware platforms to eliminate the memory bottleneck, while surpassing state-of-the-art KV cache quantization methods and achieving near-FP16 accuracy across a wide range of models.
Evaluating jailbreak attacks is challenging when prompts are not overtly harmful or fail to induce harmful outputs. Unfortunately, many existing red-teaming datasets contain such unsuitable prompts. To evaluate attacks accurately, these datasets need to be assessed and cleaned for maliciousness. However, existing malicious content detection methods rely on either manual annotation, which is labor-intensive, or large language models (LLMs), which have inconsistent accuracy in harmful types. To balance accuracy and efficiency, we propose a hybrid evaluation framework named MDH (Malicious content Detection based on LLMs with Human assistance) that combines LLM-based annotation with minimal human oversight, and apply it to dataset cleaning and detection of jailbroken responses. Furthermore, we find that well-crafted developer messages can significantly boost jailbreak success, leading us to propose two new strategies: D-Attack, which leverages context simulation, and DH-CoT, which incorporates hijacked chains of thought. The Codes, datasets, judgements, and detection results will be released in github repository: https://github.com/AlienZhang1996/DH-CoT.
Automated vulnerability detection research has made substantial progress, yet its real-world impact remains limited. Current vulnerability datasets suffer from issues including label inaccuracy rates of 20-71%, extensive duplication, and poor coverage of critical CWE types. These issues create a significant "generalization gap" where models achieve misleading self-testing performance (measured on held-out data from the same dataset for training) by exploiting spurious correlations rather than learning true vulnerability patterns. Our analysis reveals that many models experience substantial performance drops of up to 33% when evaluated on independent data, with some performing close to random guessing. To address these limitations, we present a three-part solution. First, we introduce a manually curated test dataset, BenchVul, covering the MITRE Top 25 Most Dangerous CWEs. Second, we construct a high-quality training dataset, TitanVul, comprising 38,863 functions by aggregating seven public sources and applying deduplication and validation using a novel multi-agent LLM framework. Third, we propose a Realistic Vulnerability Generation (RVG) framework, which synthesizes context-aware vulnerability examples for underrepresented but critical CWE types through simulated development workflows. Our evaluation shows the strengths of each component in closing the generalization gap. First, BenchVul shows the limitations of self-testing: models trained on existing datasets, such as BigVul and CVEfixes, experience performance drops on BenchVul (from 0.776 to 0.519 and from 0.713 to 0.607). Second, training models on TitanVul demonstrates improved generalization, with model performance increasing from 0.584 when evaluated on the same dataset to 0.767 when tested on BenchVul. Third, supplementing TitanVul with RVG-generated data yields further gains, increasing model performance by 14.0% to 0.874.
Aspect-based sentiment analysis (ABSA) has made significant strides, yet challenges remain for low-resource languages due to the predominant focus on English. Current cross-lingual ABSA studies often centre on simpler tasks and rely heavily on external translation tools. In this paper, we present a novel sequence-to-sequence method for compound ABSA tasks that eliminates the need for such tools. Our approach, which uses constrained decoding, improves cross-lingual ABSA performance by up to 10\%. This method broadens the scope of cross-lingual ABSA, enabling it to handle more complex tasks and providing a practical, efficient alternative to translation-dependent techniques. Furthermore, we compare our approach with large language models (LLMs) and show that while fine-tuned multilingual LLMs can achieve comparable results, English-centric LLMs struggle with these tasks.
We investigate the capacity of Large Language Models (LLMs) for imaginative reasoning--the proactive construction, testing, and revision of hypotheses in information-sparse environments. Existing benchmarks, often static or focused on social deduction, fail to capture the dynamic, exploratory nature of this reasoning process. To address this gap, we introduce a comprehensive research framework based on the classic "Turtle Soup" game, integrating a benchmark, an agent, and an evaluation protocol. We present TurtleSoup-Bench, the first large-scale, bilingual, interactive benchmark for imaginative reasoning, comprising 800 turtle soup puzzles sourced from both the Internet and expert authors. We also propose Mosaic-Agent, a novel agent designed to assess LLMs' performance in this setting. To evaluate reasoning quality, we develop a multi-dimensional protocol measuring logical consistency, detail completion, and conclusion alignment. Experiments with leading LLMs reveal clear capability limits, common failure patterns, and a significant performance gap compared to humans. Our work offers new insights into LLMs' imaginative reasoning and establishes a foundation for future research on exploratory agent behavior.
The rapid rise of Generative AI (GenAI), particularly LLMs, poses concerns for journalistic integrity and authorship. This study examines AI-generated content across over 40,000 news articles from major, local, and college news media, in various media formats. Using three advanced AI-text detectors (e.g., Binoculars, Fast-Detect GPT, and GPTZero), we find substantial increase of GenAI use in recent years, especially in local and college news. Sentence-level analysis reveals LLMs are often used in the introduction of news, while conclusions usually written manually. Linguistic analysis shows GenAI boosts word richness and readability but lowers formality, leading to more uniform writing styles, particularly in local media.
Spatial understanding has been a challenging task for existing Multi-modal Large Language Models~(MLLMs). Previous methods leverage large-scale MLLM finetuning to enhance MLLM's spatial understanding ability. In this paper, we present a data-efficient approach. We propose a LLM agent system with strong and advanced spatial reasoning ability, which can be used to solve the challenging spatial question answering task in complex indoor warehouse scenarios. Our system integrates multiple tools that allow the LLM agent to conduct spatial reasoning and API tools interaction to answer the given complicated spatial question. Extensive evaluations on the 2025 AI City Challenge Physical AI Spatial Intelligence Warehouse dataset demonstrate that our system achieves high accuracy and efficiency in tasks such as object retrieval, counting, and distance estimation. The code is available at: https://github.com/hsiangwei0903/SpatialAgent
Recommender systems in concert with Large Language Models (LLMs) present promising avenues for generating semantically-informed recommendations. However, LLM-based recommenders exhibit a tendency to overemphasize semantic correlations within users' interaction history. When taking pretrained collaborative ID embeddings as input, LLM-based recommenders progressively weaken the inherent collaborative signals as the embeddings propagate through LLM backbones layer by layer, as opposed to traditional Transformer-based sequential models in which collaborative signals are typically preserved or even enhanced for state-of-the-art performance. To address this limitation, we introduce FreLLM4Rec, an approach designed to balance semantic and collaborative information from a spectral perspective. Item embeddings that incorporate both semantic and collaborative information are first purified using a Global Graph Low-Pass Filter (G-LPF) to preliminarily remove irrelevant high-frequency noise. Temporal Frequency Modulation (TFM) then actively preserves collaborative signal layer by layer. Note that the collaborative preservation capability of TFM is theoretically guaranteed by establishing a connection between the optimal but hard-to-implement local graph fourier filters and the suboptimal yet computationally efficient frequency-domain filters. Extensive experiments on four benchmark datasets demonstrate that FreLLM4Rec successfully mitigates collaborative signal attenuation and achieves competitive performance, with improvements of up to 8.00\% in NDCG@10 over the best baseline. Our findings provide insights into how LLMs process collaborative information and offer a principled approach for improving LLM-based recommendation systems.
Fine-tuning large language models (LLMs) using low-rank adaptation (LoRA) has become a highly efficient approach for downstream tasks, particularly in scenarios with limited computational resources. However, applying LoRA techniques to quantized LLMs poses unique challenges due to the reduced representational precision of quantized weights. In this paper, we introduce CLoQ (Calibrated LoRA initialization for Quantized LLMs), a simplistic initialization strategy designed to overcome these challenges. Our approach focuses on minimizing the layer-wise discrepancy between the original LLM and its quantized counterpart with LoRA components during initialization. By leveraging a small calibration dataset, CLoQ quantizes a pre-trained LLM and determines the optimal LoRA components for each layer, ensuring a strong foundation for subsequent fine-tuning. A key contribution of this work is a novel theoretical result that enables the accurate and closed-form construction of these optimal LoRA components. We validate the efficacy of CLoQ across multiple tasks such as language generation, arithmetic reasoning, and commonsense reasoning, demonstrating that it consistently outperforms existing LoRA fine-tuning methods for quantized LLMs, especially at ultra low-bit widths.
The active research topic of prompt engineering makes it evident that LLMs are sensitive to small changes in prompt wording. A portion of this can be ascribed to the inductive bias that is present in the LLM. By using an LLM's output as a portion of its prompt, we can more easily create satisfactory wording for prompts. This has the effect of creating a prompt that matches the inductive bias in model. Empirically, we show that using this Inductive Bias Extraction and Matching strategy improves LLM Likert ratings used for classification by up to 19% and LLM Likert ratings used for ranking by up to 27%.
Large Language Models (LLMs) hold promise in automating data analysis tasks, yet open-source models face significant limitations in these kinds of reasoning-intensive scenarios. In this work, we investigate strategies to enhance the data analysis capabilities of open-source LLMs. By curating a seed dataset of diverse, realistic scenarios, we evaluate model behavior across three core dimensions: data understanding, code generation, and strategic planning. Our analysis reveals three key findings: (1) Strategic planning quality serves as the primary determinant of model performance; (2) Interaction design and task complexity significantly influence reasoning capabilities; (3) Data quality demonstrates a greater impact than diversity in achieving optimal performance. We leverage these insights to develop a data synthesis methodology, demonstrating significant improvements in open-source LLMs' analytical reasoning capabilities. Code is available at https://github.com/zjunlp/DataMind.
Conventional model compression techniques for LLMs address high memory consumption and slow inference challenges but typically require computationally expensive retraining to preserve accuracy. In contrast, one-shot compression methods eliminate retraining cost, but struggle to achieve accuracy comparable to dense models. This paper presents SLIM, a new one-shot compression framework that holistically integrates hardware-friendly quantization, sparsity, and low-rank approximation into a unified process. First, we formulate the quantization process using a probabilistic approach (SLIM-Quant) that enables us to apply uniform quantization. Then, we use an existing one-shot pruning method to apply semi-structured sparsity on top of the quantized weights. Finally, to compensate for the introduced aggregated quantization and sparsity error, we use a novel saliency function with unique invertible and additive features that enables us to mathematically compute the value of low-rank adapters. SLIM improves model accuracy by up to 5.66% (LLaMA-2-7B) for 2:4 sparsity with 4-bit weight quantization, outperforming prior methods. Models compressed with SLIM achieve up to 4.3x and 3.8x on Nvidia RTX3060 and A100 GPUs, respectively. Additionally, they achieve up to 0.23x end-to-end memory reduction in comparison to their dense counterparts. We also propose an optional PEFT recipe that further improves accuracy by up to 1.66% (LLaMA-2-13B) compared to SLIM without fine-tuning.
The rapid development of AI tools and implementation of LLMs within downstream tasks has been paralleled by a surge in research exploring how the outputs of such AI/LLM systems embed biases, a research topic which was already being extensively explored before the era of ChatGPT. Given the high volume of research around the biases within the outputs of AI systems and LLMs, it is imperative to conduct systematic literature reviews to document throughlines within such research. In this paper, we conduct such a review of research covering AI/LLM bias in four premier venues/organizations -- *ACL, FAccT, NeurIPS, and AAAI -- published over the past 10 years. Through a coverage of 189 papers, we uncover patterns of bias research and along what axes of human identity they commonly focus. The first emergent pattern within the corpus was that 82% (155/189) papers did not establish a working definition of "bias" for their purposes, opting instead to simply state that biases and stereotypes exist that can have harmful downstream effects while establishing only mathematical and technical definition of bias. 94 of these 155 papers have been published in the past 5 years, after Blodgett et al. (2020)'s literature review with a similar finding about NLP research and recommendation to consider how such researchers should conceptualize bias, going beyond strictly technical definitions. Furthermore, we find that a large majority of papers -- 79.9% or 151/189 papers -- focus on gender bias (mostly, gender and occupation bias) within the outputs of AI systems and LLMs. By demonstrating a strong focus within the field on gender, race/ethnicity (30.2%; 57/189), age (20.6%; 39/189), religion (19.1%; 36/189) and nationality (13.2%; 25/189) bias, we document how researchers adopt a fairly narrow conception of AI bias by overlooking several non-Western communities in fairness research, as we advocate for a stronger coverage of such populations. Finally, we note that while our corpus contains several examples of innovative debiasing methods across the aforementioned aspects of human identity, only 10.6% (20/189) include recommendations for how to implement their findings or contributions in real-world AI systems or design processes. This indicates a concerning academia-industry gap, especially since many of the biases that our corpus contains several successful mitigation methods that still persist within the outputs of AI systems and LLMs commonly used today. We conclude with recommendations towards future AI/LLM fairness research, with stronger focus on diverse marginalized populations.
Large language models (LLMs) are known to exhibit biases in downstream tasks, especially when dealing with sensitive topics such as political discourse, gender identity, ethnic relations, or national stereotypes. Although significant progress has been made in bias detection and mitigation techniques, certain challenges remain underexplored. This study proposes a reusable, granular, and topic-agnostic framework to evaluate polarisation-related biases in LLM (both open-source and closed-source). Our approach combines polarisation-sensitive sentiment metrics with a synthetically generated balanced dataset of conflict-related statements, using a predefined set of semantic categories. As a case study, we created a synthetic dataset that focusses on the Russia-Ukraine war, and we evaluated the bias in several LLMs: Llama-3, Mistral, GPT-4, Claude 3.5, and Gemini 1.0. Beyond aggregate bias scores, with a general trend for more positive sentiment toward Ukraine, the framework allowed fine-grained analysis with considerable variation between semantic categories, uncovering divergent behavioural patterns among models. Adaptation to prompt modifications showed further bias towards preconceived language and citizenship modification. Overall, the framework supports automated dataset generation and fine-grained bias assessment, is applicable to a variety of polarisation-driven scenarios and topics, and is orthogonal to many other bias-evaluation strategies.
Large language models (LLMs) have recently gained prominence in the field of software development, significantly boosting productivity and simplifying teamwork. Although prior studies have examined task-specific applications, the phase-specific effects of LLM assistance on the efficiency of code review processes remain underexplored. This research investigates the effect of GPT on GitHub pull request (PR) workflows, with a focus on reducing resolution time, optimizing phase-specific performance, and assisting developers. We curated a dataset of 25,473 PRs from 9,254 GitHub projects and identified GPT-assisted PRs using a semi-automated heuristic approach that combines keyword-based detection, regular expression filtering, and manual verification until achieving 95% labeling accuracy. We then applied statistical modeling, including multiple linear regression and Mann-Whitney U test, to evaluate differences between GPT-assisted and non-assisted PRs, both at the overall resolution level and across distinct review phases. Our research has revealed that early adoption of GPT can substantially boost the effectiveness of the PR process, leading to considerable time savings at various stages. Our findings suggest that GPT-assisted PRs reduced median resolution time by more than 60% (9 hours compared to 23 hours for non-assisted PRs). We discovered that utilizing GPT can reduce the review time by 33% and the waiting time before acceptance by 87%. Analyzing a sample dataset of 300 GPT-assisted PRs, we discovered that developers predominantly use GPT for code optimization (60%), bug fixing (26%), and documentation updates (12%). This research sheds light on the impact of the GPT model on the code review process, offering actionable insights for software teams seeking to enhance workflows and promote seamless collaboration.
Document fraud poses a significant threat to industries reliant on secure and verifiable documentation, necessitating robust detection mechanisms. This study investigates the efficacy of state-of-the-art multi-modal large language models (LLMs)-including OpenAI O1, OpenAI 4o, Gemini Flash (thinking), Deepseek Janus, Grok, Llama 3.2 and 4, Qwen 2 and 2.5 VL, Mistral Pixtral, and Claude 3.5 and 3.7 Sonnet-in detecting fraudulent documents. We benchmark these models against each other and prior work on document fraud detection techniques using a standard dataset with real transactional documents. Through prompt optimization and detailed analysis of the models' reasoning processes, we evaluate their ability to identify subtle indicators of fraud, such as tampered text, misaligned formatting, and inconsistent transactional sums. Our results reveal that top-performing multi-modal LLMs demonstrate superior zero-shot generalization, outperforming conventional methods on out-of-distribution datasets, while several vision LLMs exhibit inconsistent or subpar performance. Notably, model size and advanced reasoning capabilities show limited correlation with detection accuracy, suggesting task-specific fine-tuning is critical. This study underscores the potential of multi-modal LLMs in enhancing document fraud detection systems and provides a foundation for future research into interpretable and scalable fraud mitigation strategies.
This paper presents a novel methodology for generating synthetic Preference Optimization (PO) datasets using multi-model workflows. We evaluate the effectiveness and potential of these workflows in automating and enhancing the dataset generation process. PO dataset generation requires two modules: (1) $\textit{response evaluation}$, and (2) $\textit{response generation}$. In the $\textit{response evaluation}$ module, the responses from Large Language Models (LLMs) are evaluated and ranked - a task typically carried out by human annotators that we automate using LLMs. We assess the response evaluation module in a 2 step process. In step 1, we assess LLMs as evaluators using three distinct prompting strategies. In step 2, we apply the winning prompting strategy to compare the performance of LLM-as-a-Judge, LLMs-as-a-Jury, and LLM Debate. Our evaluation shows that GPT-4o-as-a-Judge is more consistent across all datasets. For the $\textit{response generation}$ module, we use the identified LLM evaluator configuration and compare different configurations of the LLM Feedback Loop. We use the win rate to determine the best multi-model configuration for generation. Experimenting with various configurations, we find that the LLM Feedback Loop, with Llama as the generator and Gemma as the reviewer, achieves a notable 71.8% and 73.8% win rate over single-model Llama and Gemma, respectively. After identifying the best configurations for both modules, we generate our PO datasets using the above pipeline.
By integrating the perception capabilities of multimodal encoders with the generative power of Large Language Models (LLMs), Multimodal Large Language Models (MLLMs), exemplified by GPT-4V, have achieved great success in various multimodal tasks, pointing toward a promising pathway to artificial general intelligence. Despite this progress, the limited quality of multimodal data, poor performance on many complex downstream tasks, and inadequate evaluation protocols continue to hinder the reliability and broader applicability of MLLMs across diverse domains. Inspired by the human ability to leverage external tools for enhanced reasoning and problem-solving, augmenting MLLMs with external tools (e.g., APIs, expert models, and knowledge bases) offers a promising strategy to overcome these challenges. In this paper, we present a comprehensive survey on leveraging external tools to enhance MLLM performance. Our discussion is structured along four key dimensions about external tools: (1) how they can facilitate the acquisition and annotation of high-quality multimodal data; (2) how they can assist in improving MLLM performance on challenging downstream tasks; (3) how they enable comprehensive and accurate evaluation of MLLMs; (4) the current limitations and future directions of tool-augmented MLLMs. Through this survey, we aim to underscore the transformative potential of external tools in advancing MLLM capabilities, offering a forward-looking perspective on their development and applications. The project page of this paper is publicly available athttps://github.com/Lackel/Awesome-Tools-for-MLLMs.
This work-in-progress research-to-practice paper explores the integration of Large Language Models (LLMs) into the code-review process for open-source software projects developed in computer science and software engineering courses. The focus is on developing an automated feedback tool that evaluates student code for adherence to key object-oriented design principles, addressing the need for more effective and scalable methods to teach software design best practices. The innovative practice involves leveraging LLMs and Retrieval-Augmented Generation (RAG) to create an automated feedback system that assesses student code for principles like SOLID, DRY, and design patterns. It analyzes the effectiveness of various prompting strategies and the RAG integration. Preliminary findings show promising improvements in code quality. Future work will aim to improve model accuracy and expand support for additional design principles.
Excel is a pervasive yet often complex tool, particularly for novice users, where runtime errors arising from logical mistakes or misinterpretations of functions pose a significant challenge. While large language models (LLMs) offer promising assistance by explaining formula errors, the automated correction of these semantic runtime errors remains an open problem. A primary challenge to advancing models for such scenarios is the severe lack of high-quality, comprehensive datasets for training and rigorous evaluation. This paper addresses this gap by introducing a novel approach for constructing a benchmark dataset specifically designed for Excel formula repair. We propose a data generation pipeline, which leverages a small set of curated seed samples from online forums to synthetically expand the dataset. Our pipeline integrates few-shot prompting with LLMs and employs a robust \textit{LLM-as-a-Judge} validation framework, combined with execution-based checks to ensure the correctness and semantic fidelity of the generated data. This process produced a benchmark dataset of 618 high-quality samples, covering common runtime errors. Furthermore, we propose a context-aware baseline technique for Excel formula repair that utilizes LLMs to leverage both the faulty formula, and relevant spreadsheet context. We evaluate the performance of various LLMs (GPT-4o, GPT-4.1, Phi-3, Mistral) on our newly generated benchmark using execution-based metrics. Our analysis demonstrates the dataset's quality through manual annotation and provides insights into error and function distributions. The proposed generation methodology is highly scalable and can be readily adapted to create evaluation benchmarks for similar code repair tasks in other low-resource programming languages.
Homelessness is a persistent social challenge, impacting millions worldwide. Over 770,000 people experienced homelessness in the U.S. in 2024. Social stigmatization is a significant barrier to alleviation, shifting public perception, and influencing policymaking. Given that online and city council discourse reflect and influence part of public opinion, it provides valuable insights to identify and track social biases. This research contributes to alleviating homelessness by acting on public opinion. It introduces novel methods, building on natural language processing (NLP) and large language models (LLMs), to identify and measure PEH social bias expressed in digital spaces. We present a new, manually-annotated multi-modal dataset compiled from Reddit, X (formerly Twitter), news articles, and city council meeting minutes across 10 U.S. cities. This unique dataset provides evidence of the typologies of homelessness bias described in the literature. In order to scale up and automate the detection of homelessness bias online, we evaluate LLMs as classifiers. We applied both zero-shot and few-shot classification techniques to this data. We utilized local LLMs (Llama 3.2 3B Instruct, Qwen 2.5 7B Instruct, and Phi4 Instruct Mini) as well as closed-source API models (GPT-4.1, Gemini 2.5 Pro, and Grok-4). Our findings reveal that although there are significant inconsistencies in local LLM zero-shot classification, the in-context learning classification scores of local LLMs approach the classification scores of closed-source LLMs. Furthermore, LLMs outperform BERT when averaging across all categories. This work aims to raise awareness about the pervasive bias against PEH, develop new indicators to inform policy, and ultimately enhance the fairness and ethical application of Generative AI technologies.
Large Language Models (LLMs) have significantly advanced the state-of-the-art in various coding tasks. Beyond directly answering user queries, LLMs can also serve as judges, assessing and comparing the quality of responses generated by other models. Such an evaluation capability is crucial both for benchmarking different LLMs and for improving response quality through response ranking. However, despite the growing adoption of the LLM-as-a-Judge paradigm, its effectiveness in coding scenarios remains underexplored due to the absence of dedicated benchmarks. To address this gap, we introduce CodeJudgeBench, a benchmark explicitly designed to evaluate the performance of LLM-as-a-Judge models across three critical coding tasks: code generation, code repair, and unit test generation. Through comprehensive benchmarking of 26 LLM-as-a-Judge models, we find that recent thinking models significantly outperform non-thinking models on our carefully designed code judging tasks. Notably, even relatively small thinking models, such as Qwen3-8B, can outperform specially trained LLM-as-a-Judge models up to 70B in size. Nevertheless, all models still exhibit significant randomness in their judgment of coding tasks. For pairwise judging tasks, simply changing the order in which responses are presented can substantially impact accuracy. In addition, when judging code and unit tests written by different LLMs, LLM-as-a-Judge models also show variance in performance. This sensitivity raises concerns about the reliability and consistency of LLM-as-a-Judge in coding scenarios. Lastly, we study optimal prompting strategies for LLM-as-a-Judge. We find that using pair-wise comparison outperforms scalar point-wise judging. Furthermore, retaining comments and reasoning in the full, unprocessed LLM response leads to improved judge performance.
Understanding biases and stereotypes encoded in the weights of Large Language Models (LLMs) is crucial for developing effective mitigation strategies. Biased behaviour is often subtle and non-trivial to isolate, even when deliberately elicited, making systematic analysis and debiasing particularly challenging. To address this, we introduce BiasGym, a simple, cost-effective, and generalizable framework for reliably injecting, analyzing, and mitigating conceptual associations within LLMs. BiasGym consists of two components: BiasInject, which injects specific biases into the model via token-based fine-tuning while keeping the model frozen, and BiasScope, which leverages these injected signals to identify and steer the components responsible for biased behavior. Our method enables consistent bias elicitation for mechanistic analysis, supports targeted debiasing without degrading performance on downstream tasks, and generalizes to biases unseen during token-based fine-tuning. We demonstrate the effectiveness of BiasGym in reducing real-world stereotypes (e.g., people from Italy being `reckless drivers') and in probing fictional associations (e.g., people from a fictional country having `blue skin'), showing its utility for both safety interventions and interpretability research.
The widespread deployment of LLM-based agents is likely to introduce a critical privacy threat: malicious agents that proactively engage others in multi-turn interactions to extract sensitive information. These dynamic dialogues enable adaptive attack strategies that can cause severe privacy violations, yet their evolving nature makes it difficult to anticipate and discover sophisticated vulnerabilities manually. To tackle this problem, we present a search-based framework that alternates between improving attacker and defender instructions by simulating privacy-critical agent interactions. Each simulation involves three roles: data subject, data sender, and data recipient. While the data subject's behavior is fixed, the attacker (data recipient) attempts to extract sensitive information from the defender (data sender) through persistent and interactive exchanges. To explore this interaction space efficiently, our search algorithm employs LLMs as optimizers, using parallel search with multiple threads and cross-thread propagation to analyze simulation trajectories and iteratively propose new instructions. Through this process, we find that attack strategies escalate from simple direct requests to sophisticated multi-turn tactics such as impersonation and consent forgery, while defenses advance from rule-based constraints to identity-verification state machines. The discovered attacks and defenses transfer across diverse scenarios and backbone models, demonstrating strong practical utility for building privacy-aware agents.
Amidst a shortage of qualified mental health professionals, the integration of large language models (LLMs) into psychological applications offers a promising way to alleviate the growing burden of mental health disorders. Recent reasoning-augmented LLMs have achieved remarkable performance in mathematics and programming, while research in the psychological domain has predominantly emphasized emotional support and empathetic dialogue, with limited attention to reasoning mechanisms that are beneficial to generating reliable responses. Therefore, in this paper, we propose Psyche-R1, the first Chinese psychological LLM that jointly integrates empathy, psychological expertise, and reasoning, built upon a novel data curation pipeline. Specifically, we design a comprehensive data synthesis pipeline that produces over 75k high-quality psychological questions paired with detailed rationales, generated through chain-of-thought (CoT) reasoning and iterative prompt-rationale optimization, along with 73k empathetic dialogues. Subsequently, we employ a hybrid training strategy wherein challenging samples are identified through a multi-LLM cross-selection strategy for group relative policy optimization (GRPO) to improve reasoning ability, while the remaining data is used for supervised fine-tuning (SFT) to enhance empathetic response generation and psychological domain knowledge. Extensive experiment results demonstrate the effectiveness of the Psyche-R1 across several psychological benchmarks, where our 7B Psyche-R1 achieves comparable results to 671B DeepSeek-R1.
Quantization-Aware Training (QAT) integrates quantization into the training loop, enabling LLMs to learn robust low-bit representations, and is widely recognized as one of the most promising research directions. All current QAT research focuses on minimizing quantization error on full-precision models, where the full-precision accuracy acts as an upper bound (accuracy ceiling). No existing method has even attempted to surpass this ceiling. To break this ceiling, we propose a new paradigm: raising the ceiling (full-precision model), and then still quantizing it efficiently into 2 bits. We propose Fairy$\pm i$, the first 2-bit quantization framework for complex-valued LLMs. Specifically, our method leverages the representational advantages of the complex domain to boost full-precision accuracy. We map weights to the fourth roots of unity $\{\pm1, \pm i\}$, forming a perfectly symmetric and information-theoretically optimal 2-bit representation. Importantly, each quantized weight has either a zero real or imaginary part, enabling multiplication-free inference using only additions and element swaps. Experimental results show that Fairy$\pm i$ outperforms the ceiling of existing 2-bit quantization approaches in terms of both PPL and downstream tasks, while maintaining strict storage and compute efficiency. This work opens a new direction for building highly accurate and practical LLMs under extremely low-bit constraints.
Novelty assessment is a central yet understudied aspect of peer review, particularly in high volume fields like NLP where reviewer capacity is increasingly strained. We present a structured approach for automated novelty evaluation that models expert reviewer behavior through three stages: content extraction from submissions, retrieval and synthesis of related work, and structured comparison for evidence based assessment. Our method is informed by a large scale analysis of human written novelty reviews and captures key patterns such as independent claim verification and contextual reasoning. Evaluated on 182 ICLR 2025 submissions with human annotated reviewer novelty assessments, the approach achieves 86.5% alignment with human reasoning and 75.3% agreement on novelty conclusions - substantially outperforming existing LLM based baselines. The method produces detailed, literature aware analyses and improves consistency over ad hoc reviewer judgments. These results highlight the potential for structured LLM assisted approaches to support more rigorous and transparent peer review without displacing human expertise. Data and code are made available.
Large language models (LLMs) have been widely deployed with rapidly expanding context windows to support increasingly demanding applications. However, long contexts pose significant deployment challenges, primarily due to the KV cache whose size grows proportionally with context length. While KV cache compression methods are proposed to address this issue, KV dropping methods incur considerable accuracy loss, and KV retrieval methods suffer from significant efficiency bottlenecks. We propose FreeKV, an algorithm-system co-optimization framework to enhance KV retrieval efficiency while preserving accuracy. On the algorithm side, FreeKV introduces speculative retrieval to shift the KV selection and recall processes out of the critical path, combined with fine-grained correction to ensure accuracy. On the system side, FreeKV employs hybrid KV layouts across CPU and GPU memory to eliminate fragmented data transfers, and leverages double-buffered streamed recall to further improve efficiency. Experiments demonstrate that FreeKV achieves near-lossless accuracy across various scenarios and models, delivering up to 13$\times$ speedup compared to SOTA KV retrieval methods.
Large language models are often assumed to acquire increasingly structured, generalizable internal representations simply by scaling data and parameters. We interrogate this assumption by introducing a Clinical Trial Natural Language Inference benchmark comprising four reasoning families, Causal Attribution, Compositional Grounding, Epistemic Verification, and Risk State Abstraction. Each item is paired with a targeted Ground Knowledge and Meta-Level Reasoning Verification (GKMRV) probe, allowing us to dissociate failures of factual access from failures of inference. We evaluate six contemporary LLMs under both direct and chain of thought prompting. Models achieve near-ceiling GKMRV accuracy (mean accuracy 0.918) yet perform poorly on the main reasoning tasks (mean accuracy 0.25). Despite low accuracy, output inferences are highly consistent across samples (mean 0.87), indicating a systematic application of underlying heuristics and shortcuts. These results reveal fundamental structural and representational limitations: current LLMs often possess the relevant clinical knowledge but lack the structured, composable internal representations needed to deploy it reliably (e.g., integrating constraints, weighing evidence, or simulating counterfactuals). Decoupling knowledge from reasoning with GKMRV makes this dissociation explicit and measurable, providing an effective framework for probing the reliability of LLMs in high-stakes domains.
The Key-Value (KV) cache, which stores intermediate attention computations (Key and Value pairs) to avoid redundant calculations, is a fundamental mechanism for accelerating Large Language Model (LLM) inference. However, this efficiency optimization introduces significant yet underexplored privacy risks. This paper provides the first comprehensive analysis of these vulnerabilities, demonstrating that an attacker can reconstruct sensitive user inputs directly from the KV-cache. We design and implement three distinct attack vectors: a direct Inversion Attack, a more broadly applicable and potent Collision Attack, and a semantic-based Injection Attack. These methods demonstrate the practicality and severity of KV-cache privacy leakage issues. To mitigate this, we propose KV-Cloak, a novel, lightweight, and efficient defense mechanism. KV-Cloak uses a reversible matrix-based obfuscation scheme, combined with operator fusion, to secure the KV-cache. Our extensive experiments show that KV-Cloak effectively thwarts all proposed attacks, reducing reconstruction quality to random noise. Crucially, it achieves this robust security with virtually no degradation in model accuracy and minimal performance overhead, offering a practical solution for trustworthy LLM deployment.
Recommendation systems often suffer from data sparsity caused by limited user-item interactions, which degrade their performance and amplify popularity bias in real-world scenarios. This paper proposes a novel data augmentation framework that leverages Large Language Models (LLMs) and item textual descriptions to enrich interaction data. By few-shot prompting LLMs multiple times to rerank items and aggregating the results via majority voting, we generate high-confidence synthetic user-item interactions, supported by theoretical guarantees based on the concentration of measure. To effectively leverage the augmented data in the context of a graph recommendation system, we integrate it into a graph contrastive learning framework to mitigate distributional shift and alleviate popularity bias. Extensive experiments show that our method improves accuracy and reduces popularity bias, outperforming strong baselines.
The Object Goal Navigation (ObjectNav) task challenges agents to locate a specified object in an unseen environment by imagining unobserved regions of the scene. Prior approaches rely on deterministic and discriminative models to complete semantic maps, overlooking the inherent uncertainty in indoor layouts and limiting their ability to generalize to unseen environments. In this work, we propose GOAL, a generative flow-based framework that models the semantic distribution of indoor environments by bridging observed regions with LLM-enriched full-scene semantic maps. During training, spatial priors inferred from large language models (LLMs) are encoded as two-dimensional Gaussian fields and injected into target maps, distilling rich contextual knowledge into the flow model and enabling more generalizable completions. Extensive experiments demonstrate that GOAL achieves state-of-the-art performance on MP3D and Gibson, and shows strong generalization in transfer settings to HM3D. Codes and pretrained models are available at https://github.com/Badi-Li/GOAL.
Large language models (LLMs) have revolutionized natural language processing, yet their tendency to hallucinate poses serious challenges for reliable deployment. Despite numerous hallucination detection methods, their evaluations often rely on ROUGE, a metric based on lexical overlap that misaligns with human judgments. Through comprehensive human studies, we demonstrate that while ROUGE exhibits high recall, its extremely low precision leads to misleading performance estimates. In fact, several established detection methods show performance drops of up to 45.9\% when assessed using human-aligned metrics like LLM-as-Judge. Moreover, our analysis reveals that simple heuristics based on response length can rival complex detection techniques, exposing a fundamental flaw in current evaluation practices. We argue that adopting semantically aware and robust evaluation frameworks is essential to accurately gauge the true performance of hallucination detection methods, ultimately ensuring the trustworthiness of LLM outputs.
Quantifying the influence of individual training samples is essential for enhancing the transparency and accountability of large language models (LLMs) and vision-language models (VLMs). However, existing data valuation methods often rely on Hessian information or model retraining, making them computationally prohibitive for billion-parameter models. In this work, we introduce For-Value, a forward-only data valuation framework that enables scalable and efficient influence estimation for both LLMs and VLMs. By leveraging the rich representations of modern foundation models, For-Value computes influence scores using a simple closed-form expression based solely on a single forward pass, thereby eliminating the need for costly gradient computations. Our theoretical analysis demonstrates that For-Value accurately estimates per-sample influence by capturing alignment in hidden representations and prediction errors between training and validation samples. Extensive experiments show that For-Value matches or outperforms gradient-based baselines in identifying impactful fine-tuning examples and effectively detecting mislabeled data.
Large language models (LLMs) excel at solving problems with clear and complete statements, but often struggle with nuanced environments or interactive tasks which are common in most real-world scenarios. This highlights the critical need for developing LLMs that can effectively engage in logically consistent multi-turn dialogue, seek information and reason with incomplete data. To this end, we introduce a novel benchmark comprising a suite of multi-turn tasks each designed to test specific reasoning, interactive dialogue, and information-seeking abilities. These tasks have deterministic scoring mechanisms, thus eliminating the need for human intervention. Evaluating frontier models on our benchmark reveals significant headroom. Our analysis shows that most errors emerge from poor instruction following, reasoning failures, and poor planning. This benchmark provides valuable insights into the strengths and weaknesses of current LLMs in handling complex, interactive scenarios and offers a robust platform for future research aimed at improving these critical capabilities.
Semantic conflicts arise when a developer introduces changes to a codebase that unintentionally affect the behavior of changes integrated in parallel by other developers. Traditional merge tools are unable to detect such conflicts, so complementary tools like SMAT have been proposed. SMAT relies on generating and executing unit tests: if a test fails on the base version, passes on a developer's modified version, but fails again after merging with another developer's changes, a semantic conflict is indicated. While SMAT is effective at detecting conflicts, it suffers from a high rate of false negatives, partly due to the limitations of unit test generation tools such as Randoop and Evosuite. To investigate whether large language models (LLMs) can overcome these limitations, we propose and integrate a new test generation tool based on Code Llama 70B into SMAT. We explore the model's ability to generate tests using different interaction strategies, prompt contents, and parameter configurations. Our evaluation uses two samples: a benchmark with simpler systems from related work, and a more significant sample based on complex, real-world systems. We assess the effectiveness of the new SMAT extension in detecting conflicts. Results indicate that, although LLM-based test generation remains challenging and computationally expensive in complex scenarios, there is promising potential for improving semantic conflict detection.
For large language models (LLMs), reasoning over graphs could help solve many problems. Prior work has tried to improve LLM graph reasoning by examining how best to serialize graphs as text and by combining GNNs and LLMs. However, the merits of such approaches remain unclear, so we empirically answer the following research questions: (1) Can LLMs learn to solve fundamental graph tasks without specialized graph encoding models?, (2) Can LLMs generalize learned solutions to unseen graph structures or tasks?, and (3) What are the merits of competing approaches to learn graph tasks? We show that even small LLMs can learn to solve graph tasks by training them with instructive chain-of-thought solutions, and this training generalizes, without specialized graph encoders, to new tasks and graph structures.
Large language models (LLMs) have shown promising performance across diverse domains. Many practical applications of LLMs, such as code completion and structured data extraction, require adherence to syntactic constraints specified by a formal language. Yet, due to their probabilistic nature, LLM output is not guaranteed to adhere to such formal languages. Prior work has proposed constrained decoding as a means to restrict LLM generation to particular formal languages. However, existing works are not applicable to the emerging paradigm of diffusion LLMs, when used in practical scenarios such as the generation of formally correct C++ or JSON output. In this paper we address this challenge and present the first constrained decoding method for diffusion models, one that can handle formal languages captured by context-free grammars. We begin by reducing constrained decoding to the more general additive infilling problem, which asks whether a partial output can be completed to a valid word in the target language. This problem also naturally subsumes the previously unaddressed multi-region infilling constrained decoding. We then reduce this problem to the task of deciding whether the intersection of the target language and a regular language is empty and present an efficient algorithm to solve it for context-free languages. Empirical results on various applications, such as C++ code infilling and structured data extraction in JSON, demonstrate that our method achieves near-perfect syntactic correctness while consistently preserving or improving functional correctness. Importantly, our efficiency optimizations ensure that the computational overhead remains practical.
Evaluating AI agents within complex, interactive environments that mirror real-world challenges is critical for understanding their practical capabilities. While existing agent benchmarks effectively assess skills like tool use or performance on structured tasks, they often do not fully capture an agent's ability to operate autonomously in exploratory environments that demand sustained, self-directed reasoning over a long and growing context. To enable a more accurate assessment of AI agents in challenging exploratory environments, we introduce TextQuests, a benchmark based on the Infocom suite of interactive fiction games. These text-based adventures, which can take human players over 30 hours and require hundreds of precise actions to solve, serve as an effective proxy for evaluating AI agents on focused, stateful tasks. The benchmark is specifically designed to assess an LLM agent's capacity for self-contained problem-solving by precluding the use of external tools, thereby focusing on intrinsic long-context reasoning capabilities in an exploratory environment characterized by the need for trial-and-error learning and sustained problem-solving within a single interactive session. We release TextQuests at https://textquests.ai.
Traditional end-of-quarter surveys often fail to provide instructors with timely, detailed, and actionable feedback about their teaching. In this paper, we explore how Large Language Model (LLM)-powered chatbots can reimagine the classroom feedback process by engaging students in reflective, conversational dialogues. Through the design and deployment of a three-part system-PromptDesigner, FeedbackCollector, and FeedbackAnalyzer-we conducted a pilot study across two graduate courses at UC Santa Cruz. Our findings suggest that LLM-based feedback systems offer richer insights, greater contextual relevance, and higher engagement compared to standard survey tools. Instructors valued the system's adaptability, specificity, and ability to support mid-course adjustments, while students appreciated the conversational format and opportunity for elaboration. We conclude by discussing the design implications of using AI to facilitate more meaningful and responsive feedback in higher education.
With the increasing popularity of large language models (LLMs) for a variety of tasks, there has been a growing interest in strategies that can predict which out of a set of LLMs will yield a successful answer at low cost. This problem promises to become more and more relevant as providers like Microsoft allow users to easily create custom LLM "assistants" specialized to particular types of queries. However, some tasks (i.e., queries) may be too specialized and difficult for a single LLM to handle alone. These applications often benefit from breaking down the task into smaller subtasks, each of which can then be executed by a LLM expected to perform well on that specific subtask. For example, in extracting a diagnosis from medical records, one can first select an LLM to summarize the record, select another to validate the summary, and then select another, possibly different, LLM to extract the diagnosis from the summarized record. Unlike existing LLM selection or routing algorithms, this setting requires that we select a sequence of LLMs, with the output of each LLM feeding into the next and potentially influencing its success. Thus, unlike single LLM selection, the quality of each subtask's output directly affects the inputs, and hence the cost and success rate, of downstream LLMs, creating complex performance dependencies that must be learned and accounted for during selection. We propose a neural contextual bandit-based algorithm that trains neural networks that model LLM success on each subtask in an online manner, thus learning to guide the LLM selections for the different subtasks, even in the absence of historical LLM performance data. Experiments on telecommunications question answering and medical diagnosis prediction datasets illustrate the effectiveness of our proposed approach compared to other LLM selection algorithms.
This paper presents Block, a distributed scheduling framework designed to optimize load balancing and auto-provisioning across instances in large language model serving frameworks by leveraging contextual information from incoming requests. Unlike popular model serving systems that rely on monolithic and heuristic task schedulers, Block operates as a fully distributed, stateless, and predictive scheduling system to achieve low overhead, reliability, and scalability. It leverages the deterministic and predictable characteristics of LLM inferences, such as host configurations, response lengths, and hardware performance, to make scheduling decisions based on accurately predicted metrics. Evaluation on a 12 GPUs cluster shows that Block significantly outperforms heuristic schedulers, boosting serving capacity by up to 16.7\% and reducing P99 tail latency by up to 49.5\%. These performance gains remain consistent across diverse models, workloads and configurations. Code and data are open-sourced.
As Large Language Models (LLMs) become increasingly integrated into real-world applications, ensuring their outputs align with human values and safety standards has become critical. The field has developed diverse alignment approaches including traditional fine-tuning methods (RLHF, instruction tuning), post-hoc correction systems, and inference-time interventions, each with distinct advantages and limitations. However, the lack of unified evaluation frameworks makes it difficult to systematically compare these paradigms and guide deployment decisions. This paper introduces a multi-dimensional evaluation of alignment techniques for LLMs, a comprehensive evaluation framework that provides a systematic comparison across all major alignment paradigms. Our framework assesses methods along four key dimensions: alignment detection, alignment quality, computational efficiency, and robustness. Through experiments across diverse base models and alignment strategies, we demonstrate the utility of our framework in identifying strengths and limitations of current state-of-the-art models, providing valuable insights for future research directions.
Automated vulnerability detection research has made substantial progress, yet its real-world impact remains limited. Current vulnerability datasets suffer from issues including label inaccuracy rates of 20-71%, extensive duplication, and poor coverage of critical CWE types. These issues create a significant "generalization gap" where models achieve misleading self-testing performance (measured on held-out data from the same dataset for training) by exploiting spurious correlations rather than learning true vulnerability patterns. Our analysis reveals that many models experience substantial performance drops of up to 33% when evaluated on independent data, with some performing close to random guessing. To address these limitations, we present a three-part solution. First, we introduce a manually curated test dataset, BenchVul, covering the MITRE Top 25 Most Dangerous CWEs. Second, we construct a high-quality training dataset, TitanVul, comprising 38,863 functions by aggregating seven public sources and applying deduplication and validation using a novel multi-agent LLM framework. Third, we propose a Realistic Vulnerability Generation (RVG) framework, which synthesizes context-aware vulnerability examples for underrepresented but critical CWE types through simulated development workflows. Our evaluation shows the strengths of each component in closing the generalization gap. First, BenchVul shows the limitations of self-testing: models trained on existing datasets, such as BigVul and CVEfixes, experience performance drops on BenchVul (from 0.776 to 0.519 and from 0.713 to 0.607). Second, training models on TitanVul demonstrates improved generalization, with model performance increasing from 0.584 when evaluated on the same dataset to 0.767 when tested on BenchVul. Third, supplementing TitanVul with RVG-generated data yields further gains, increasing model performance by 14.0% to 0.874.
Data annotation underpins the success of modern AI, but the aggregation of crowd-collected datasets can harm the preservation of diverse perspectives in data. Difficult and ambiguous tasks cannot easily be collapsed into unitary labels. Prior work has shown that deliberation and discussion improve data quality and preserve diverse perspectives -- however, synchronous deliberation through crowdsourcing platforms is time-intensive and costly. In this work, we create a Socratic dialog system using Large Language Models (LLMs) to act as a deliberation partner in place of other crowdworkers. Against a benchmark of synchronous deliberation on two tasks (Sarcasm and Relation detection), our Socratic LLM encouraged participants to consider alternate annotation perspectives, update their labels as needed (with higher confidence), and resulted in higher annotation accuracy (for the Relation task where ground truth is available). Qualitative findings show that our agent's Socratic approach was effective at encouraging reasoned arguments from our participants, and that the intervention was well-received. Our methodology lays the groundwork for building scalable systems that preserve individual perspectives in generating more representative datasets.
Forecasting in real-world settings requires models to integrate not only historical data but also relevant contextual information, often available in textual form. While recent work has shown that large language models (LLMs) can be effective context-aided forecasters via na\"ive direct prompting, their full potential remains underexplored. We address this gap with 4 strategies, providing new insights into the zero-shot capabilities of LLMs in this setting. ReDP improves interpretability by eliciting explicit reasoning traces, allowing us to assess the model's reasoning over the context independently from its forecast accuracy. CorDP leverages LLMs solely to refine existing forecasts with context, enhancing their applicability in real-world forecasting pipelines. IC-DP proposes embedding historical examples of context-aided forecasting tasks in the prompt, substantially improving accuracy even for the largest models. Finally, RouteDP optimizes resource efficiency by using LLMs to estimate task difficulty, and routing the most challenging tasks to larger models. Evaluated on different kinds of context-aided forecasting tasks from the CiK benchmark, our strategies demonstrate distinct benefits over na\"ive prompting across LLMs of different sizes and families. These results open the door to further simple yet effective improvements in LLM-based context-aided forecasting.
Multimodal LLMs (MLLMs) have reached remarkable levels of proficiency in understanding multimodal inputs. However, understanding and interpreting the behavior of such complex models is a challenging task, not to mention the dynamic shifts that may occur during fine-tuning, or due to covariate shift between datasets. In this work, we apply concept-level analysis towards MLLM understanding. More specifically, we propose to map hidden states to interpretable visual and textual concepts. This enables us to more efficiently compare certain semantic dynamics, such as the shift from an original and fine-tuned model, revealing concept alteration and potential biases that may occur during fine-tuning. We also demonstrate the use of shift vectors to capture these concepts changes. These shift vectors allow us to recover fine-tuned concepts by applying simple, computationally inexpensive additive concept shifts in the original model. Finally, our findings also have direct applications for MLLM steering, which can be used for model debiasing as well as enforcing safety in MLLM output. All in all, we propose a novel, training-free, ready-to-use framework for MLLM behavior interpretability and control. Our implementation is publicly available.
Recent studies have raised significant concerns regarding the reliability of current mathematics benchmarks, highlighting issues such as simplistic design and potential data contamination. Consequently, developing a reliable benchmark that effectively evaluates large language models' (LLMs) genuine capabilities in mathematical reasoning remains a critical challenge. To address these concerns, we propose RV-Bench, a novel evaluation methodology for Benchmarking LLMs with Random Variables in mathematical reasoning. Specifically, we build question-generating functions to produce random variable questions (RVQs), whose background content mirrors original benchmark problems, but with randomized variable combinations, rendering them "unseen" to LLMs. Models must completely understand the inherent question pattern to correctly answer RVQs with diverse variable combinations. Thus, an LLM's genuine reasoning capability is reflected through its accuracy and robustness on RV-Bench. We conducted extensive experiments on over 30 representative LLMs across more than 1,000 RVQs. Our findings propose that LLMs exhibit a proficiency imbalance between encountered and ``unseen'' data distributions. Furthermore, RV-Bench reveals that proficiency generalization across similar mathematical reasoning tasks is limited, but we verified it can still be effectively elicited through test-time scaling.
In this paper, we propose LibRec, a novel framework that integrates the capabilities of LLMs with retrieval-augmented generation(RAG) techniques to automate the recommendation of alternative libraries. The framework further employs in-context learning to extract migration intents from commit messages to enhance the accuracy of its recommendations. To evaluate the effectiveness of LibRec, we introduce LibEval, a benchmark designed to assess the performance in the library migration recommendation task. LibEval comprises 2,888 migration records associated with 2,368 libraries extracted from 2,324 Python repositories. Each migration record captures source-target library pairs, along with their corresponding migration intents and intent types. Based on LibEval, we evaluated the effectiveness of ten popular LLMs within our framework, conducted an ablation study to examine the contributions of key components within our framework, explored the impact of various prompt strategies on the framework's performance, assessed its effectiveness across various intent types, and performed detailed failure case analyses.
In the rapidly evolving field of Explainable Natural Language Processing (NLP), textual explanations, i.e., human-like rationales, are pivotal for explaining model predictions and enriching datasets with interpretable labels. Traditional approaches rely on human annotation, which is costly, labor-intensive, and impedes scalability. In this work, we present an automated framework that leverages multiple state-of-the-art large language models (LLMs) to generate high-quality textual explanations. We rigorously assess the quality of these LLM-generated explanations using a comprehensive suite of Natural Language Generation (NLG) metrics. Furthermore, we investigate the downstream impact of these explanations on the performance of pre-trained language models (PLMs) and LLMs across natural language inference tasks on two diverse benchmark datasets. Our experiments demonstrate that automated explanations exhibit highly competitive effectiveness compared to human-annotated explanations in improving model performance. Our findings underscore a promising avenue for scalable, automated LLM-based textual explanation generation for extending NLP datasets and enhancing model performance.
We present a novel approach to bias mitigation in large language models (LLMs) by applying steering vectors to modify model activations in forward passes. We compute 8 steering vectors, each corresponding to a different social bias axis, such as age, gender, or race, on a training subset of the BBQ dataset and compare the effectiveness of these to 3 additional bias mitigation methods across 4 datasets. When optimized on the BBQ dataset, our individually tuned steering vectors achieve average improvements of 12.8% on BBQ, 8.3% on CLEAR-Bias, and 1% on StereoSet, and show improvements over prompting and Self-Debias in all cases, and improvements over fine-tuning in 12 out of 17 evaluations. In addition, steering vectors showed the lowest impact on MMLU scores of the four bias mitigation methods tested. The work presents the first systematic investigation of steering vectors for bias mitigation, and we demonstrate that they are a powerful and computationally efficient strategy for reducing bias in LLMs, with broader implications for enhancing AI safety.
Mental health disorders are increasingly prevalent worldwide, creating an urgent need for innovative tools to support early diagnosis and intervention. This study explores the potential of Large Language Models (LLMs) in multimodal mental health diagnostics, specifically for detecting depression and Post Traumatic Stress Disorder through text and audio modalities. Using the E-DAIC dataset, we compare text and audio modalities to investigate whether LLMs can perform equally well or better with audio inputs. We further examine the integration of both modalities to determine if this can enhance diagnostic accuracy, which generally results in improved performance metrics. Our analysis specifically utilizes custom-formulated metrics; Modal Superiority Score and Disagreement Resolvement Score to evaluate how combined modalities influence model performance. The Gemini 1.5 Pro model achieves the highest scores in binary depression classification when using the combined modality, with an F1 score of 0.67 and a Balanced Accuracy (BA) of 77.4%, assessed across the full dataset. These results represent an increase of 3.1% over its performance with the text modality and 2.7% over the audio modality, highlighting the effectiveness of integrating modalities to enhance diagnostic accuracy. Notably, all results are obtained in zero-shot inferring, highlighting the robustness of the models without requiring task-specific fine-tuning. To explore the impact of different configurations on model performance, we conduct binary, severity, and multiclass tasks using both zero-shot and few-shot prompts, examining the effects of prompt variations on performance. The results reveal that models such as Gemini 1.5 Pro in text and audio modalities, and GPT-4o mini in the text modality, often surpass other models in balanced accuracy and F1 scores across multiple tasks.
Pairwise evaluation of Large Language Models (LLMs) is a common paradigm, but it is prone to preference bias, where judges systematically favor certain outputs, such as their own. This bias leads to inconsistent and skewed rankings across different judges. To address this, we first empirically demonstrate significant and heterogeneous biases in cross-model evaluations. We then propose UDA (Unsupervised Debiasing Alignment), a framework that reduces inter-judge disagreement by dynamically adjusting the Elo rating system. For each pairwise comparison, a compact neural network learns to adaptively set the K-factor and refine win probabilities. Crucially, UDA operates in a fully unsupervised manner, guided solely by the objective of minimizing the dispersion among the Elo trajectories of all judges. This forces an alignment towards a collective consensus, which serves as an unsupervised proxy for a more stable and reproducible evaluation. In addition, we provide theoretical motivation demonstrating how alignment towards a consensus can reduce aggregate system bias. Experiments show that UDA significantly reduces the inter-judge rating standard deviation by up to 63.4% and improves the average correlation with human judgments by 24.7%. Notably, UDA elevates the performance of poorly performing judges to achieve parity with high-quality ones, fostering a more robust and reliable evaluation ecosystem. Code and data are available at https://anonymous.4open.science/r/62AB93CD-23B4.
Human speech goes beyond the mere transfer of information; it is a profound exchange of emotions and a connection between individuals. While Text-to-Speech (TTS) models have made huge progress, they still face challenges in controlling the emotional expression in the generated speech. In this work, we propose EmoVoice, a novel emotion-controllable TTS model that exploits large language models (LLMs) to enable fine-grained freestyle natural language emotion control, and a phoneme boost variant design that makes the model output phoneme tokens and audio tokens in parallel to enhance content consistency, inspired by chain-of-thought (CoT) and chain-of-modality (CoM) techniques. Besides, we introduce EmoVoice-DB, a high-quality 40-hour English emotion dataset featuring expressive speech and fine-grained emotion labels with natural language descriptions. EmoVoice achieves state-of-the-art performance on the English EmoVoice-DB test set using only synthetic training data, and on the Chinese Secap test set using our in-house data. We further investigate the reliability of existing emotion evaluation metrics and their alignment with human perceptual preferences, and explore using SOTA multimodal LLMs GPT-4o-audio and Gemini to assess emotional speech. Dataset, code, checkpoints, and demo samples are available at https://github.com/yanghaha0908/EmoVoice.
As large language models (LLMs) grow more capable, they face growing vulnerability to sophisticated jailbreak attacks. While developers invest heavily in alignment finetuning and safety guardrails, researchers continue publishing novel attacks, driving progress through adversarial iteration. This dynamic mirrors a strategic game of continual evolution. However, two major challenges hinder jailbreak development: the high cost of querying top-tier LLMs and the short lifespan of effective attacks due to frequent safety updates. These factors limit cost-efficiency and practical impact of research in jailbreak attacks. To address this, we propose MetaCipher, a low-cost, multi-agent jailbreak framework that generalizes across LLMs with varying safety measures. Using reinforcement learning, MetaCipher is modular and adaptive, supporting extensibility to future strategies. Within as few as 10 queries, MetaCipher achieves state-of-the-art attack success rates on recent malicious prompt benchmarks, outperforming prior jailbreak methods. We conduct a large-scale empirical evaluation across diverse victim models and benchmarks, demonstrating its robustness and adaptability. Warning: This paper contains model outputs that may be offensive or harmful, shown solely to demonstrate jailbreak efficacy.
Design studies aim to create visualization solutions for real-world problems of different application domains. Recently, the emergence of large language models (LLMs) has introduced new opportunities to enhance the design study process, providing capabilities such as creative problem-solving, data handling, and insightful analysis. However, despite their growing popularity, there remains a lack of systematic understanding of how LLMs can effectively assist researchers in visualization-specific design studies. In this paper, we conducted a multi-stage qualitative study to fill this gap, involving 30 design study researchers from diverse backgrounds and expertise levels. Through in-depth interviews and carefully-designed questionnaires, we investigated strategies for utilizing LLMs, the challenges encountered, and the practices used to overcome them. We further compiled and summarized the roles that LLMs can play across different stages of the design study process. Our findings highlight practical implications to inform visualization practitioners, and provide a framework for leveraging LLMs to enhance the design study process in visualization research.
Large Language Models (LLMs) have recently demonstrated strong capabilities in translating natural language into database queries, especially when dealing with complex graph-structured data. However, real-world queries often contain inherent ambiguities, and the interconnected nature of graph structures can amplify these challenges, leading to unintended or incorrect query results. To systematically evaluate LLMs on this front, we propose a taxonomy of graph-query ambiguities, comprising three primary types: Attribute Ambiguity, Relationship Ambiguity, and Attribute-Relationship Ambiguity, each subdivided into Same-Entity and Cross-Entity scenarios. We introduce AmbiGraph-Eval, a novel benchmark of real-world ambiguous queries paired with expert-verified graph query answers. Evaluating 9 representative LLMs shows that even top models struggle with ambiguous graph queries. Our findings reveal a critical gap in ambiguity handling and motivate future work on specialized resolution techniques.
Large language models (LLMs) hold promise in clinical decision support but face major challenges in safety evaluation and effectiveness validation. We developed the Clinical Safety-Effectiveness Dual-Track Benchmark (CSEDB), a multidimensional framework built on clinical expert consensus, encompassing 30 criteria covering critical areas like critical illness recognition, guideline adherence, and medication safety, with weighted consequence measures. Thirty-two specialist physicians developed and reviewed 2,069 open-ended Q&A items aligned with these criteria, spanning 26 clinical departments to simulate real-world scenarios. Benchmark testing of six LLMs revealed moderate overall performance (average total score 57.2%, safety 54.7%, effectiveness 62.3%), with a significant 13.3% performance drop in high-risk scenarios (p < 0.0001). Domain-specific medical LLMs showed consistent performance advantages over general-purpose models, with relatively higher top scores in safety (0.912) and effectiveness (0.861). The findings of this study not only provide a standardized metric for evaluating the clinical application of medical LLMs, facilitating comparative analyses, risk exposure identification, and improvement directions across different scenarios, but also hold the potential to promote safer and more effective deployment of large language models in healthcare environments.
Chain-of-Thought (CoT) prompting has been shown to improve Large Language Model (LLM) performance on various tasks. With this approach, LLMs appear to produce human-like reasoning steps before providing answers (a.k.a., CoT reasoning), which often leads to the perception that they engage in deliberate inferential processes. However, some initial findings suggest that CoT reasoning may be more superficial than it appears, motivating us to explore further. In this paper, we study CoT reasoning via a data distribution lens and investigate if CoT reasoning reflects a structured inductive bias learned from in-distribution data, allowing the model to conditionally generate reasoning paths that approximate those seen during training. Thus, its effectiveness is fundamentally bounded by the degree of distribution discrepancy between the training data and the test queries. With this lens, we dissect CoT reasoning via three dimensions: task, length, and format. To investigate each dimension, we design DataAlchemy, an isolated and controlled environment to train LLMs from scratch and systematically probe them under various distribution conditions. Our results reveal that CoT reasoning is a brittle mirage that vanishes when it is pushed beyond training distributions. This work offers a deeper understanding of why and when CoT reasoning fails, emphasizing the ongoing challenge of achieving genuine and generalizable reasoning.
This technical report accompanies the LLM robustness leaderboard published by PRISM Eval for the Paris AI Action Summit. We introduce PRISM Eval Behavior Elicitation Tool (BET), an AI system performing automated red-teaming through Dynamic Adversarial Optimization that achieves 100% Attack Success Rate (ASR) against 37 of 41 state-of-the-art LLMs. Beyond binary success metrics, we propose a fine-grained robustness metric estimating the average number of attempts required to elicit harmful behaviors, revealing that attack difficulty varies by over 300-fold across models despite universal vulnerability. We introduce primitive-level vulnerability analysis to identify which jailbreaking techniques are most effective for specific hazard categories. Our collaborative evaluation with trusted third parties from the AI Safety Network demonstrates practical pathways for distributed robustness assessment across the community.
Modern computing systems, such as HDFS and Spark, produce vast quantities of logs that developers use for tasks like anomaly detection and error analysis. To simplify log analysis, template generation methods have been proposed to standardize log formats, transforming unstructured data into structured templates. Existing heuristic-based methods and neural network-based methods suffer from low accuracy problems due to the reliance on handcrafted heuristics or specific log patterns in training sets. Recently, large language models (LLMs) have shown great potential in log template generation. However, they often struggle with ambiguous, complex, or highly specific log content, which can lead to errors in generating accurate templates. To address these challenges, we propose LLMLog, a multi-round annotation framework with adaptive in-context learning. We first propose an edit-distance-based similarity metric to evaluate log similarity. Then, we introduce a method to select the most informative $k$ unlabeled logs for annotation by considering both the representativeness of the logs and the confidence of LLM predictions. Additionally, we design an adaptive context selection strategy that adaptively selects labeled logs to ensure comprehensive keyword coverage for unlabeled logs. These labeled logs serve as the context for LLMs to better understand the unlabeled logs, thereby enhancing the accuracy of template generation. Extensive experiments on sixteen datasets demonstrate that LLMLog outperforms the state-of-the-art approaches.
Log statements have become an integral part of modern software systems. Prior research efforts have focused on supporting the decisions of placing log statements, such as where/what to log. With the increasing adoption of Large Language Models (LLMs) for code-related tasks such as code completion or generation, automated approaches for generating log statements have gained much momentum. However, the performance of these approaches still has a long way to go. This paper explores enhancing the performance of LLM-based solutions for automated log statement generation by post-training LLMs with a purpose-built dataset. Thus the primary contribution is a novel approach called AUCAD, which automatically constructs such a dataset with information extracting from log-related issues. Researchers have long noticed that a significant portion of the issues in the open-source community are related to log statements. However, distilling this portion of data requires manual efforts, which is labor-intensive and costly, rendering it impractical. Utilizing our approach, we automatically extract log-related issues from 1,537 entries of log data across 88 projects and identify 808 code snippets (i.e., methods) with retrievable source code both before and after modification of each issue (including log statements) to construct a dataset. Each entry in the dataset consists of a data pair representing high-quality and problematic log statements, respectively. With this dataset, we proceed to post-train multiple LLMs (primarily from the Llama series) for automated log statement generation. Both human and experimental evaluations indicate that these models significantly outperform existing LLM-based solutions, thereby validating the efficacy of our method for constructing a post-training dataset to enhance LLM-based log statement generation.
We introduce MapStory, an LLM-powered animation prototyping tool that generates editable map animation sequences directly from natural language text by leveraging a dual-agent LLM architecture. Given a user written script, MapStory automatically produces a scene breakdown, which decomposes the text into key map animation primitives such as camera movements, visual highlights, and animated elements. Our system includes a researcher agent that accurately queries geospatial information by leveraging an LLM with web search, enabling automatic extraction of relevant regions, paths, and coordinates while allowing users to edit and query for changes or additional information to refine the results. Additionally, users can fine-tune parameters of these primitive blocks through an interactive timeline editor. We detail the system's design and architecture, informed by formative interviews with professional animators and by an analysis of 200 existing map animation videos. Our evaluation, which includes expert interviews (N=5) and a usability study (N=12), demonstrates that MapStory enables users to create map animations with ease, facilitates faster iteration, encourages creative exploration, and lowers barriers to creating map-centric stories.
Large Language Models (LLMs) have revolutionized content creation across digital platforms, offering unprecedented capabilities in natural language generation and understanding. These models enable beneficial applications such as content generation, question and answering (Q&A), programming, and code reasoning. Meanwhile, they also pose serious risks by inadvertently or intentionally producing toxic, offensive, or biased content. This dual role of LLMs, both as powerful tools for solving real-world problems and as potential sources of harmful language, presents a pressing sociotechnical challenge. In this survey, we systematically review recent studies spanning unintentional toxicity, adversarial jailbreaking attacks, and content moderation techniques. We propose a unified taxonomy of LLM-related harms and defenses, analyze emerging multimodal and LLM-assisted jailbreak strategies, and assess mitigation efforts, including reinforcement learning with human feedback (RLHF), prompt engineering, and safety alignment. Our synthesis highlights the evolving landscape of LLM safety, identifies limitations in current evaluation methodologies, and outlines future research directions to guide the development of robust and ethically aligned language technologies.
Large Language Models (LLMs) have demonstrated remarkable capabilities in natural language processing tasks, yet their application to graph structure analysis, particularly in community search, remains underexplored. Community search, a fundamental task in graph analysis, aims to identify groups of nodes with dense interconnections, which is crucial for understanding the macroscopic structure of graphs. In this paper, we propose GraphCS, a comprehensive benchmark designed to evaluate the performance of LLMs in community search tasks. Our experiments reveal that while LLMs exhibit preliminary potential, they frequently fail to return meaningful results and suffer from output bias. To address these limitations, we introduce CS-Agent, a dual-agent collaborative framework to enhance LLM-based community search. CS-Agent leverages the complementary strengths of two LLMs acting as Solver and Validator. Through iterative feedback and refinement, CS-Agent dynamically refines initial results without fine-tuning or additional training. After the multi-round dialogue, Decider module selects the optimal community. Extensive experiments demonstrate that CS-Agent significantly improves the quality and stability of identified communities compared to baseline methods. To our knowledge, this is the first work to apply LLMs to community search, bridging the gap between LLMs and graph analysis while providing a robust and adaptive solution for real-world applications.