Large language models (LLMs) are known to have the potential to generate harmful content, posing risks to users. While significant progress has been made in developing taxonomies for LLM risks and safety evaluation prompts, most studies have focused on monolingual contexts, primarily in English. However, language- and region-specific risks in bilingual contexts are often overlooked, and core findings can diverge from those in monolingual settings. In this paper, we introduce Qorǵau, a novel dataset specifically designed for safety evaluation in Kazakh and Russian, reflecting the unique bilingual context in Kazakhstan, where both Kazakh (a low-resource language) and Russian (a high-resource language) are spoken. Experiments with both multilingual and language-specific LLMs reveal notable differences in safety performance, emphasizing the need for tailored, region-specific datasets to ensure the responsible and safe deployment of LLMs in countries like Kazakhstan. Warning: this paper contains example data that may be offensive, harmful, or biased.
论文检索
输入标题、作者或关键词，从 3,086 篇学术成果中精准定位
Identifying bias in LLM-generated content is a crucial prerequisite for ensuring fairness in LLMs. Existing methods, such as fairness classifiers and LLM-based judges, face limitations related to difficulties in understanding underlying intentions and the lack of criteria for fairness judgment. In this paper, we introduce BiasGuard, a novel bias detection tool that explicitly analyzes inputs and reasons through fairness specifications to provide accurate judgments. BiasGuard is implemented through a two-stage approach: the first stage initializes the model to explicitly reason based on fairness specifications, while the second stage leverages reinforcement learning to enhance its reasoning and judgment capabilities. Our experiments, conducted across five datasets, demonstrate that BiasGuard outperforms existing tools, improving accuracy and reducing over-fairness misjudgments. We also highlight the importance of reasoning-enhanced decision-making and provide evidence for the effectiveness of our two-stage optimization pipeline.
The rapid proliferation of fake news across multiple domains poses significant threats to society. Existing multi-domain detection models typically capture domain-shared semantic features to achieve generalized detection. However, they often fail to generalize well due to poor adaptability, which limits their ability to provide complementary features for detection, especially in data-constrained conditions. To address these challenges, we investigate the propagation-adaptive multi-domain fake news detection paradigm. We propose a novel framework, Structure-adaptive Adversarial Contrastive Learning (StruACL), to adaptively enable structure knowledge transfer between multiple domains. Specifically, we first contrast representations between content-only and propagation-rich data to preserve structural patterns in the shared representation space. Additionally, we design a propagation-guided adversarial training strategy to enhance the diversity of representations. Under the StruACL objective, we leverage a unified Transformer-based and graph-based model to jointly learn transferable semantic and structural features for detection across multiple domains. Experiments on seven fake news datasets demonstrate that StruACL-TGN achieves better multi-domain detection performance on general and data-constrained scenarios, showing the effectiveness and better generalization of StruACL.
Random Splitting Negatively Impacts NER Evaluation: Quantifying and Eliminating the Overestimation of NER Performance
PDF ↗In named entity recognition (NER), models are evaluated on their ability to identify entity mentions in text. However, standard evaluation methods often rely on test sets that contain named entities already present in the training data, raising concerns about overestimation of model performance.This work investigates the impact of varying degrees of entity contamination on a dataset level on the generalization ability and reported F1 scores of three state-of-the-art NER models.Experiments on five standard benchmarks show that F1 scores for contaminated entities statistically significantly inflate reported F1 scores as contamination rates increase, with F1 performance gaps ranging from 2-10% compared to entities not seen during training.To address these inflated F1 scores, we additionally propose a novel NER dataset splitting method using a minimum cut algorithm to minimize train-test entity leakage.While our splitting method ensures near-zero entity contamination, we also compare new and existing dataset splits on named entity sample counts.
The continuous evolution and enhanced reasoning capabilities of large language models (LLMs) have elevated their role in complex tasks, notably in travel planning, where demand for personalized, high-quality itineraries is rising. However, current benchmarks often rely on unrealistic simulated data, failing to reflect the differences between LLM-generated and real-world itineraries. Existing evaluation metrics, which primarily emphasize constraints, fall short of providing a comprehensive assessment of the overall quality of travel plans. To address these limitations, we introduce TripTailor, a benchmark designed specifically for personalized travel planning in real-world scenarios. This dataset features an extensive collection of over 500,000 real-world points of interest (POIs) and nearly 4,000 diverse travel itineraries, complete with detailed information, providing a more authentic evaluation framework. Experiments show that fewer than 10% of the itineraries generated by the latest state-of-the-art LLMs achieve human-level performance. Moreover, we identify several critical challenges in travel planning, including the feasibility, rationality, and personalized customization of the proposed solutions. We hope that TripTailor will drive the development of travel planning agents capable of understanding and meeting user needs while generating practical itineraries.
Large language models (LLMs) have revolutionized the field of natural language processing, enabling remarkable progress in various tasks. Different from objective tasks such as commonsense reasoning and arithmetic question-answering, the performance of LLMs on subjective tasks is still limited, where the perspective on the specific problem plays crucial roles for better interpreting the context and giving proper response. For example, in certain scenarios, LLMs may perform better when answering from an expert role perspective, potentially eliciting their relevant domain knowledge. In contrast, in some scenarios, LLMs may provide more accurate responses when answering from a third-person standpoint, enabling a more comprehensive understanding of the problem and potentially mitigating inherent biases. In this paper, we propose Reasoning through Perspective Transition (RPT), a method based on in-context learning that enables LLMs to dynamically select among direct, role, and third-person perspectives for the best way to solve corresponding subjective problem. Through extensive experiments on totally 12 subjective tasks by using both closed-source and open-source LLMs including GPT-4, GPT-3.5, Llama-3, and Qwen-2, our method outperforms widely used single fixed perspective based methods such as chain-of-thought prompting and expert prompting, highlights the intricate ways that LLMs can adapt their perspectives to provide nuanced and contextually appropriate responses for different problems.
Visual agent models for automating human activities on Graphical User Interfaces (GUIs) have emerged as a promising research direction, driven by advances in large Vision Language Models (VLMs). A critical challenge in GUI automation is the precise grounding of interface elements across diverse platforms. Existing vision-only GUI agents directly ground elements from large and cluttered screenshots, requiring them to process substantial irrelevant information that compromises their accuracy. In addition, these approaches typically employ basic cross-entropy loss for learning grounding objectives, which fails to effectively capture grounding quality compared to established object detection metrics like Intersection-over-Union (IoU). To address these issues, we introduce R-VLM, a novel GUI grounding approach that leverages zoomed-in region proposals for precise element localization. We also propose an IoU-aware objective function that facilitates model convergence toward high IoU predictions. Our approach bridges the gap between VLMs and conventional object detection techniques, improving the state-of-the-art grounding accuracy by 13% across diverse GUI platforms on the GUI grounding benchmarks ScreenSpot and AgentStudio. In addition, our R-VLM approach shows 3.2-9.7% absolute accuracy improvements in GUI navigation tasks on the AITW and Mind2Web benchmarks.
Large Vision-Language Models (LVLMs) have recently gained attention due to their distinctive performance and broad applicability. While it has been previously shown that their efficacy in usage scenarios involving non-Western contexts falls short, existing studies are limited in scope, covering just a narrow range of cultures, focusing exclusively on a small number of cultural aspects, or evaluating a limited selection of models on a single task only. Towards globally inclusive LVLM research, we introduce GIMMICK, an extensive multimodal benchmark designed to assess a broad spectrum of cultural knowledge across 144 countries representing six global macro-regions. GIMMICK comprises six tasks built upon three new datasets that span 728 unique cultural events or facets on which we evaluated 20 LVLMs and 11 LLMs, including five proprietary and 26 open-weight models of all sizes. We systematically examine (1) regional cultural biases, (2) the influence of model size, (3) input modalities, and (4) external cues. Our analyses reveal strong biases toward Western cultures across models and tasks and highlight strong correlations between model size and performance, as well as the effectiveness of multimodal input and external geographic cues. We further find that models have more knowledge of tangible than intangible aspects (e.g., food vs. rituals) and that they excel in recognizing broad cultural origins but struggle with a more nuanced understanding.
Supervised and Unsupervised Probing of Shortcut Learning: Case Study on the Emergence and Evolution of Syntactic Heuristics in BERT
PDF ↗Contemporary language models (LMs) such as BERT (Devlin et al., 2019, T5 (Raffel et al., 2023), GPT-4 (OpenAI, 2023), have exhibited remarkable capabilities, effectively addressing long-standing challenges in the field. However, these models rely on shortcut learning, using a decision rule that relies on superficial cues that are spuriously correlated with the labels (Geirhos et al., 2020). In this research, we focus on the reliance on a specific type of shortcuts, namely syntactic heuristics, in BERT when performing Natural Language Inference (NLI), a representative task in Natural Language Understanding (Jeretic et al., 2020). By making use of two probing methods, one supervised, one unsupervised, we investigate where these shortcuts emerge, how they evolve and how they impact the latent knowledge of the LM. Our findings reveal that syntactic heuristics are absent in pretrained models but emerge and evolve as the model is finetuned with datasets of increasing size. The adoption of these shortcuts varies across different hidden layers, with specific layers closer to the output contributing more to this phenomenon. Despite the model’s reliance on shortcuts during inference, it retains information relevant to the task, and our supervised and unsupervised probes process this information differently.
Focused-DPO: Enhancing Code Generation Through Focused Preference Optimization on Error-Prone Points
PDF ↗Code generation models have shown significant potential for automating programming tasks. However, the challenge of generating accurate and reliable code persists due to the highly complex and long-reasoning nature of the task. Even state-of-the-art models often fail in code generation due to small errors, which can drastically affect the overall functionality of code. Our study identifies that current models tend to produce errors concentrated at specific error-prone points, which significantly impacts the accuracy of the generated code. To address this issue, we introduce Focused-DPO, a framework that enhances code generation by directing preference optimization towards these critical error-prone areas. This approach builds on Direct Preference Optimization, emphasizing accuracy in parts prone to errors. Additionally, we develop a method called Error-Point Identification, which constructs a dataset that targets these problematic points without requiring costly human annotations. Our experiments on benchmarks such as HumanEval(+), MBPP(+), and LiveCodeBench demonstrate that Focused-DPO significantly improves the precision and reliability of code generation, reducing common errors and enhancing overall code quality. By focusing on error-prone points, Focused-DPO advances the accuracy and functionality of model-generated code.
Debt Collection Negotiations with Large Language Models: An Evaluation System and Optimizing Decision Making with Multi-Agent
PDF ↗Debt collection negotiations (DCN) are vital for managing non-performing loans (NPLs) and reducing creditor losses. Traditional methods are labor-intensive, while large language models (LLMs) offer promising automation potential. However, prior systems lacked dynamic negotiation and real-time decision-making capabilities. This paper explores LLMs in automating DCN and proposes a novel evaluation framework with 13 metrics across 4 aspects. Our experiments reveal that LLMs tend to over-concede compared to human negotiators. To address this, we propose the Multi-Agent Debt Negotiation (MADeN) framework, incorporating planning and judging modules to improve decision rationality. We also apply post-training techniques, including DPO with rejection sampling, to optimize performance. Our studies provide valuable insights for practitioners and researchers seeking to enhance efficiency and outcomes in this domain.
Dealing with long and highly complex technical text is a challenge for Large Language Models (LLMs), which still have to unfold their potential in supporting expensive and time intensive processes like patent drafting. Within patents, the description constitutes more than 90% of the document on average. Yet, its automatic generation remains understudied. When drafting patent applications, patent attorneys typically receive invention reports (IRs), which are usually confidential, hindering research on LLM-supported patent drafting.Often, pre-publication research papers serve as IRs. We leverage this duality to build PAP2PAT, an open and realistic benchmark for patent drafting consisting of 1.8k patent-paper pairs describing the same inventions. To address the complex long-document patent generation task, we propose chunk-based outline-guided generation using the research paper as invention specification. Our extensive evaluation using PAP2PAT and a human case study show that LLMs can effectively leverage information from the paper, but still struggle to provide the necessary level of detail. Fine-tuning leads to more patent-style language, but also to more hallucination. We release our data and code.
Nowadays, single Large Language Model (LLM) struggles with critical issues such as hallucination and inadequate reasoning abilities. To mitigate these issues, Multi-Agent Debate (MAD) has emerged as an effective strategy, where LLM agents engage in in-depth debates with others on tasks. However, existing MAD methods face two major issues: (a) too lengthy input contexts, which causes LLM agents to get lost in plenty of input information and experiences performance drop; and (b) the overconfidence dilemma, where self-assured LLM agents dominate the debate, leading to low debating effectiveness. To address these limitations, we propose a novel MAD method called ”CortexDebate”. Inspired by the human brain’s tendency to establish a sparse and dynamically optimized network among cortical areas governed by white matter, CortexDebate constructs a sparse debating graph among LLM agents, where each LLM agent only debates with the ones that are helpful to it. To optimize the graph, we propose a module named McKinsey-based Debate Matter (MDM), which acts as an artificial analog to white matter. By integrating the McKinsey Trust Formula, a well-established measure of trustworthiness from sociology, MDM enables credible evaluations that guide graph optimization. The effectiveness of our CortexDebate has been well demonstrated by extensive experimental results across eight datasets from four task types.
Large Language Models (LLMs) are being used more and more extensively for automated evaluation in various scenarios. Previous studies have attempted to fine-tune open-source LLMs to replicate the evaluation explanations and judgments of powerful proprietary models, such as GPT-4. However, these methods are largely limited to text-based analyses under predefined general criteria, resulting in reduced adaptability for unseen instructions and demonstrating instability in evaluating adherence to quantitative and structural constraints. To address these limitations, we propose a novel evaluation framework, ARJudge, that adaptively formulates evaluation criteria and synthesizes both text-based and code-driven analyses to evaluate LLM responses. ARJudge consists of two components: a fine-tuned Analyzer that generates multi-faceted evaluation analyses and a tuning-free Refiner that combines and refines all analyses to make the final judgment. We construct a Composite Analysis Corpus that integrates tasks for evaluation criteria generation alongside text-based and code-driven analysis generation to train the Analyzer. Our results demonstrate that ARJudge outperforms existing fine-tuned evaluators in effectiveness and robustness. Furthermore, it demonstrates the importance of multi-faceted evaluation and code-driven analyses in enhancing evaluation capabilities.
Test-time compute is emerging as a new paradigm for enhancing language models’ complex multi-step reasoning capabilities, as demonstrated by the success of OpenAI’s o1 and o3, as well as DeepSeek’s R1. Compared to explicit reasoning in test-time compute, implicit reasoning is more inference-efficient, requiring fewer generated tokens. However, why does the advanced reasoning capability fail to emerge in the implicit reasoning style? In this work, we train GPT-2 from scratch on a curated multi-step mathematical reasoning dataset and conduct analytical experiments to investigate how language models perform implicit reasoning in multi-step tasks. Our findings reveal: 1) Language models can perform step-by-step reasoning and achieve high accuracy in both in-domain and out-of-domain tests via implicit reasoning. However, this capability only emerges when trained on fixed-pattern data. 2) Conversely, implicit reasoning abilities emerging from training on unfixed-pattern data tend to overfit a specific pattern and fail to generalize further. Notably, this limitation is also observed in state-of-the-art large language models. These findings suggest that language models acquire implicit reasoning through shortcut learning, enabling strong performance on tasks with similar patterns while lacking generalization. Resources are available at https://github.com/TianheL/LM-Implicit-Reasoning.
Inverse tasks can uncover potential reasoning gaps as Large Language Models (LLMs) scale up. In this work, we explore the redefinition task, in which we assign alternative values to well-known physical constants and units of measure, prompting LLMs to respond accordingly. Our findings show that not only does model performance degrade with scale, but its false confidence also rises. Moreover, while factors such as prompting strategies or response formatting are influential, they do not preclude LLMs from anchoring to memorized values.
EC-RAFT: Automated Generation of Clinical Trial Eligibility Criteria through Retrieval-Augmented Fine-Tuning
PDF ↗Eligibility criteria (EC) are critical components of clinical trial design, defining the parameters for participant inclusion and exclusion. However, designing EC remains a complex, expertise-intensive process. Traditional approaches to EC generation may fail to produce comprehensive, contextually appropriate criteria. To address these challenges, we introduce EC-RAFT, a method that utilizes Retrieval-Augmented Fine-Tuning (RAFT) to generate structured and cohesive EC directly from clinical trial titles and descriptions. EC-RAFT integrates contextual retrieval, synthesized intermediate reasoning, and fine-tuned language models to produce comprehensive EC sets. To enhance clinical alignment evaluation with referenced criteria, we also propose an LLM-guided evaluation pipeline. Our results demonstrate that our solution, which uses Llama-3.1-8B-Instruct as a base model, achieves a BERTScore of 86.23 and an EC-matched LLM-as-a-Judge score of 1.66 out of 3, outperforming zero-shot Llama-3.1 and Gemini-1.5 by 0.41 and 0.11 points, respectively. On top of that, EC-RAFT also outperforms other fine-tuned versions of Llama-3.1. EC-RAFT was trained in a low-cost setup and, therefore, can be used as a practical solution for EC generation while ensuring quality and relevance in clinical trial design. We release our code on GitHub at https://github.com/biodatlab/ec-raft/
The field of neural machine translation (NMT) has changed with the advent of large language models (LLMs). Much of the recent emphasis in natural language processing (NLP) has been on modeling machine translation and many other problems using a single pre-trained Transformer decoder, while encoder-decoder architectures, which were the standard in earlier NMT models, have received relatively less attention. In this paper, we explore translation models that are universal, efficient, and easy to optimize, by marrying the world of LLMs with the world of NMT. We apply LLMs to NMT encoding and leave the NMT decoder unchanged. We also develop methods for adapting LLMs to work better with the NMT decoder. Furthermore, we construct a new dataset involving multiple tasks to assess how well the machine translation system generalizes across various tasks. Evaluations on the WMT and our datasets show that results using our method match or surpass a range of baselines in terms of translation quality, but achieve 2.4 \sim 6.5 \times inference speedups and a 75% reduction in the memory footprint of the KV cache. It also demonstrates strong generalization across a variety of translation-related tasks.
FinRipple: Aligning Large Language Models with Financial Market for Event Ripple Effect Awareness
PDF ↗Financial markets exhibit complex dynamics where localized events trigger ripple effects across entities. Previous event studies, constrained by static single-companies analyses and simplistic assumptions, fail to capture these ripple effects. While large language models (LLMs) offer emergent reasoning capabilities, their direct application falters due to structural market unawareness and limited capacity to analyze ripple effects. We propose FinRipple, an elegant framework that empowers LLMs with the ability to analyze ripple effects through financial theory-guided large-scale reinforcement learning. We begin by relaxing the assumptions of previous methods, incorporating a time-varying knowledge graph to accurately represent market structure. By seamlessly integrating classical asset pricing theory, we align the LLM with the market, enabling it to predict ripple effects. To the best of our knowledge, we are the first to provide a standardized definition of ripple effect prediction, a task that is extremely important yet unexplored in the financial domain. Extensive experiments demonstrate that FinRipple provides a promising solution to this task.
In recent years, researchers have started analyzing the cultural sensitivity of LLMs. In this respect, Etiquettes have been an active area of research. Etiquettes are region-specific and are an essential part of the culture of a region; hence, it is imperative to make LLMs sensitive to etiquettes. However, there needs to be more resources in evaluating LLMs for their understanding and bias with regard to etiquettes. In this resource paper, we introduce EtiCor++, a corpus of etiquettes worldwide. We introduce different tasks for evaluating LLMs for knowledge about etiquettes across various regions. Further, we introduce various metrics for measuring bias in LLMs. Extensive experimentation with LLMs shows inherent bias towards certain regions.