Recently, building retrieval-augmented generation (RAG) systems to enhance the capability of large language models (LLMs) has become a common practice. Especially in the legal domain, previous judicial decisions play a significant role under the doctrine of stare decisis which emphasizes the importance of making decisions based on (retrieved) prior documents. However, the overall performance of RAG system depends on many components: (1) retrieval corpora, (2) retrieval algorithms, (3) rerankers, (4) LLM backbones, and (5) evaluation metrics. Here we propose LRAGE, an open-source tool for holistic evaluation of RAG systems focusing on the legal domain. LRAGE provides GUI and CLI interfaces to facilitate seamless experiments and investigate how changes in the aforementioned five components affect the overall accuracy. We validated LRAGE using multilingual legal benches including Korean (KBL), English (LegalBench), and Chinese (LawBench) by demonstrating how the overall accuracy changes when varying the five components mentioned above. The source code is available at https://github.com/hoorangyee/LRAGE.
Evaluating large language models (LLMs) effectively remains a critical bottleneck, as traditional static benchmarks suffer from saturation and contamination, while human evaluations are costly and slow. This hinders timely or domain-specific assessment, crucial for real-world applications. We introduce YourBench, a novel, open-source framework that addresses these limitations by enabling dynamic, automated generation of reliable, up-to-date, and domain-tailored benchmarks cheaply and without manual annotation, directly from user-provided documents. We demonstrate its efficacy by replicating 7 diverse MMLU subsets using minimal source text, achieving this for under 15 USD in total inference costs while perfectly preserving the relative model performance rankings (Spearman Rho = 1) observed on the original benchmark. To ensure that YourBench generates data grounded in provided input instead of relying on posterior parametric knowledge in models, we also introduce Tempora-0325, a novel dataset of over 7K diverse documents, published exclusively after March 2025. Our comprehensive analysis spans 26 SoTA models from 7 major families across varying scales (3-671B parameters) to validate the quality of generated evaluations through rigorous algorithmic checks (e.g., citation grounding) and human assessments. We release the YourBench library, the Tempora-0325 dataset, 150k+ question answer pairs based on Tempora and all evaluation and inference traces to facilitate reproducible research and empower the community to generate bespoke benchmarks on demand, fostering more relevant and trustworthy LLM evaluation.
Large language models (LLMs) exhibit remarkable multilingual capabilities despite the extreme language imbalance in the pre-training data. In this paper, we closely examine the reasons behind this phenomenon, focusing on the pre-training corpus. We find that the existence of code-switching, alternating between different languages within a context, is key to multilingual capabilities. We conduct an analysis to investigate code-switching in the pre-training corpus, examining its presence and categorizing it into four types within two quadrants. We then assess its impact on multilingual performance. These types of code-switching data are unbalanced in proportions and demonstrate different effects on facilitating language transfer. To better explore the power of code-switching for language alignment during pre-training, we investigate the strategy of synthetic code-switching. We continuously scale up the synthetic code-switching data and observe remarkable improvements in both benchmarks and representation space. Extensive experiments indicate that incorporating synthetic code-switching data enables better language alignment and generalizes well to high, medium, and low-resource languages with pre-training corpora of varying qualities.
Theory of Mind (ToM), the ability to attribute mental states to others, is fundamental for human social intelligence and a critical capability for advanced Artificial Intelligence. Recent advancements in Large Language Models (LLMs) have shown promising performance on ToM benchmarks, raising the question: Do these benchmarks necessitate explicit human-like reasoning processes, or can models succeed through alternative strategies? We investigate this question empirically by applying Reinforcement Learning (RL) and Supervised Fine-Tuning (SFT) to LLMs of varying scales (0.5B to 7B parameters) and evaluating them across multiple ToM datasets. Our results reveal a scale-dependent impact of RL: while RL significantly improves accuracy and fosters high-quality, interpretable, and transferable belief-tracking reasoning in larger models (7B), it leads to "reasoning collapse" in smaller models ($\leq$3B), where high accuracy and generalization ability are achieved via drastically shortened, less meaningful responses. Surprisingly, further SFT achieves competitive and generalizable performance across these benchmarks, often matching or exceeding RL models in accuracy, despite not being explicitly trained to produce structured reasoning traces. These findings highlight a critical discrepancy between benchmark accuracy and the nature of learned reasoning. Our work suggests that current ToM benchmarks may be solvable without requiring the explicit, human-like simulation of mental states they were designed to probe. LLMs, particularly when scale is limited or training signals focus solely on output correctness, may leverage alternative rules effective for benchmark data structures.
The best performing transformer-based language models use subword tokenization techniques, such as Byte-Pair-Encoding (BPE). However, these approaches often overlook linguistic principles, such as morphological segmentation, which we believe is fundamental for understanding language-specific word structure. In this study, we leverage an annotated Danish morphological dataset to train a semisupervised model for morphological segmentation, enabling the development of tokenizers optimized for Danish morphology. We evaluate four distinct tokenizers, including two custom morphological tokenizers, by analyzing their performance in morphologically segmenting Danish words. Additionally, we train two generative transformer models, \textit{CerebrasGPT-111M} and \textit{LLaMA-3.2 1B}, using these tokenizers and evaluate their downstream performance. Our findings reveal that our custom-developed tokenizers substantially enhance morphological segmentation, achieving an F1 score of 58.84, compared to 39.28 achieved by a Danish BPE tokenizer. In downstream tasks, models trained with our morphological tokenizers outperform those using BPE tokenizers across different evaluation metrics. These results highlight that incorporating Danish morphological segmentation strategies into tokenizers leads to improved performance in generative transformer models on Danish language
The detrimental effects of toxicity in competitive online video games are widely acknowledged, prompting publishers to monitor player chat conversations. This is challenging due to the context-dependent nature of toxicity, often spread across multiple messages or informed by non-textual interactions. Traditional toxicity detectors focus on isolated messages, missing the broader context needed for accurate moderation. This is especially problematic in video games, where interactions involve specialized slang, abbreviations, and typos, making it difficult for standard models to detect toxicity, especially given its rarity. We adapted RoBERTa LLM to support moderation tailored to video games, integrating both textual and non-textual context. By enhancing pretrained embeddings with metadata and addressing the unique slang and language quirks through domain adaptive pretraining, our method better captures the nuances of player interactions. Using two gaming datasets - from Defense of the Ancients 2 (DOTA 2) and Call of Duty$^\circledR$: Modern Warfare$^\circledR$III (MWIII) we demonstrate which sources of context (metadata, prior interactions...) are most useful, how to best leverage them to boost performance, and the conditions conducive to doing so. This work underscores the importance of context-aware and domain-specific approaches for proactive moderation.
Predicting future events stands as one of the ultimate aspirations of artificial intelligence. Recent advances in large language model (LLM)-based systems have shown remarkable potential in forecasting future events, thereby garnering significant interest in the research community. Currently, several benchmarks have been established to evaluate the forecasting capabilities by formalizing the event prediction as a retrieval-augmented generation (RAG) and reasoning task. In these benchmarks, each prediction question is answered with relevant retrieved news articles. However, because there is no consideration on whether the questions can be supported by valid or sufficient supporting rationales, some of the questions in these benchmarks may be inherently noninferable. To address this issue, we introduce a new benchmark, PROPHET, which comprises inferable forecasting questions paired with relevant news for retrieval. To ensure the inferability of the benchmark, we propose Causal Intervened Likelihood (CIL), a statistical measure that assesses inferability through causal inference. In constructing this benchmark, we first collected recent trend forecasting questions and then filtered the data using CIL, resulting in an inferable benchmark for event prediction. Through extensive experiments, we first demonstrate the validity of CIL and in-depth investigations into event prediction with the aid of CIL. Subsequently, we evaluate several representative prediction systems on PROPHET, drawing valuable insights for future directions.
In an era where AI-driven hiring is transforming recruitment practices, concerns about fairness and bias have become increasingly important. To explore these issues, we introduce a benchmark, FAIRE (Fairness Assessment In Resume Evaluation), to test for racial and gender bias in large language models (LLMs) used to evaluate resumes across different industries. We use two methods-direct scoring and ranking-to measure how model performance changes when resumes are slightly altered to reflect different racial or gender identities. Our findings reveal that while every model exhibits some degree of bias, the magnitude and direction vary considerably. This benchmark provides a clear way to examine these differences and offers valuable insights into the fairness of AI-based hiring tools. It highlights the urgent need for strategies to reduce bias in AI-driven recruitment. Our benchmark code and dataset are open-sourced at our repository: https://github.com/athenawen/FAIRE-Fairness-Assessment-In-Resume-Evaluation.git.
This paper presents LITE, an LLM-based evaluation method designed for efficient and flexible assessment of taxonomy quality. To address challenges in large-scale taxonomy evaluation, such as efficiency, fairness, and consistency, LITE adopts a top-down hierarchical evaluation strategy, breaking down the taxonomy into manageable substructures and ensuring result reliability through cross-validation and standardized input formats. LITE also introduces a penalty mechanism to handle extreme cases and provides both quantitative performance analysis and qualitative insights by integrating evaluation metrics closely aligned with task objectives. Experimental results show that LITE demonstrates high reliability in complex evaluation tasks, effectively identifying semantic errors, logical contradictions, and structural flaws in taxonomies, while offering directions for improvement. Code is available at https://github.com/Zhang-l-i-n/TAXONOMY_DETECT .
The newly released OpenAI-o1 and DeepSeek-R1 have demonstrated that test-time scaling can significantly improve model performance, especially in complex tasks such as logical reasoning. Common test-time scaling methods involve generating more chain of thoughts (CoTs) or longer CoTs with self-correction. However, while self-correction can improve performance, it may lead to significant token waste and reduce readability of the CoT if the reasoning steps are already correct. To demonstrate that large language models (LLMs) can rectify errors at a more fine-grained level, we propose Adaptive Rectification Sampling (AR-Sampling), which can guide the LLMs to self-correction at the appropriate step. AR-Sampling leverages a process-supervised reward model (PRM) as a verifier and constructed trigger sentences to guide the model in adaptive step-level rethinking. Through the experiments on GSM8K and MATH500, it indicate that our approach enables the models to rethink in more fine-grained level, improving the accuracy of solutions, while generating a reasonable number of additional tokens.
We present ThinkPrune, a simple yet effective method for pruning the thinking length for long-thinking LLMs, which has been found to often produce inefficient and redundant thinking processes. Existing preliminary explorations of reducing thinking length primarily focus on forcing the thinking process to early exit, rather than adapting the LLM to optimize and consolidate the thinking process, and therefore the length-performance tradeoff observed so far is sub-optimal. To fill this gap, ThinkPrune offers a simple solution that continuously trains the long-thinking LLMs via reinforcement learning (RL) with an added token limit, beyond which any unfinished thoughts and answers will be discarded, resulting in a zero reward. To further preserve model performance, we introduce an iterative length pruning approach, where multiple rounds of RL are conducted, each with an increasingly more stringent token limit. We observed that ThinkPrune results in a remarkable performance-length tradeoff -- on the AIME24 dataset, the reasoning length of DeepSeek-R1-Distill-Qwen-1.5B can be reduced by half with only 2% drop in performance. We also observed that after pruning, the LLMs can bypass unnecessary steps while keeping the core reasoning process complete. Code is available at https://github.com/UCSB-NLP-Chang/ThinkPrune.
Text-to-Image (T2I) models often suffer from issues such as semantic leakage, incorrect feature binding, and omissions of key concepts in the generated image. This work studies these phenomena by looking into the role of information flow between textual token representations. To this end, we generate images by applying the diffusion component on a subset of contextual token representations in a given prompt and observe several interesting phenomena. First, in many cases, a word or multiword expression is fully represented by one or two tokens, while other tokens are redundant. For example, in "San Francisco's Golden Gate Bridge", the token "gate" alone captures the full expression. We demonstrate the redundancy of these tokens by removing them after textual encoding and generating an image from the resulting representation. Surprisingly, we find that this process not only maintains image generation performance but also reduces errors by 21\% compared to standard generation. We then show that information can also flow between different expressions in a sentence, which often leads to semantic leakage. Based on this observation, we propose a simple, training-free method to mitigate semantic leakage: replacing the leaked item's representation after the textual encoding with its uncontextualized representation. Remarkably, this simple approach reduces semantic leakage by 85\%. Overall, our work provides a comprehensive analysis of information flow across textual tokens in T2I models, offering both novel insights and practical benefits.
Cultural Intelligence (CQ) refers to the ability to understand unfamiliar cultural contexts-a crucial skill for large language models (LLMs) to effectively engage with globally diverse users. While existing research often focuses on explicitly stated cultural norms, such approaches fail to capture the subtle, implicit values that underlie real-world conversations. To address this gap, we introduce CQ-Bench, a benchmark specifically designed to assess LLMs' capability to infer implicit cultural values from natural conversational contexts. We generate a multi-character conversation-based stories dataset using values from the World Value Survey and GlobalOpinions datasets, with topics including ethical, religious, social, and political. Our dataset construction pipeline includes rigorous validation procedures-incorporation, consistency, and implicitness checks-using GPT-4o, with 98.2% human-model agreement in the final validation. Our benchmark consists of three tasks of increasing complexity: attitude detection, value selection, and value extraction. We find that while o1 and Deepseek-R1 models reach human-level performance in value selection (0.809 and 0.814), they still fall short in nuanced attitude detection, with F1 scores of 0.622 and 0.635, respectively. In the value extraction task, GPT-4o-mini and o3-mini score 0.602 and 0.598, highlighting the difficulty of open-ended cultural reasoning. Notably, fine-tuning smaller models (e.g., LLaMA-3.2-3B) on only 500 culturally rich examples improves performance by over 10%, even outperforming stronger baselines (o3-mini) in some cases. Using CQ-Bench, we provide insights into the current challenges in LLMs' CQ research and suggest practical pathways for enhancing LLMs' cross-cultural reasoning abilities.
Computer use agents automate digital tasks by directly interacting with graphical user interfaces (GUIs) on computers and mobile devices, offering significant potential to enhance human productivity by completing an open-ended space of user queries. However, current agents face significant challenges: imprecise grounding of GUI elements, difficulties with long-horizon task planning, and performance bottlenecks from relying on single generalist models for diverse cognitive tasks. To this end, we introduce Agent S2, a novel compositional framework that delegates cognitive responsibilities across various generalist and specialist models. We propose a novel Mixture-of-Grounding technique to achieve precise GUI localization and introduce Proactive Hierarchical Planning, dynamically refining action plans at multiple temporal scales in response to evolving observations. Evaluations demonstrate that Agent S2 establishes new state-of-the-art (SOTA) performance on three prominent computer use benchmarks. Specifically, Agent S2 achieves 18.9% and 32.7% relative improvements over leading baseline agents such as Claude Computer Use and UI-TARS on the OSWorld 15-step and 50-step evaluation. Moreover, Agent S2 generalizes effectively to other operating systems and applications, surpassing previous best methods by 52.8% on WindowsAgentArena and by 16.52% on AndroidWorld relatively. Code available at https://github.com/simular-ai/Agent-S.
This paper presents a simple, effective, and cost-efficient strategy to improve LLM performance by scaling test-time compute. Our strategy builds upon the repeated-sampling-then-voting framework, with a novel twist: incorporating multiple models, even weaker ones, to leverage their complementary strengths that potentially arise from diverse training data and paradigms. By using consistency as a signal, our strategy dynamically switches between models. Theoretical analysis highlights the efficiency and performance advantages of our strategy. Extensive experiments on six datasets demonstrate that our strategy not only outperforms self-consistency and state-of-the-art multi-agent debate approaches, but also significantly reduces inference costs. Additionally, ModelSwitch requires only a few comparable LLMs to achieve optimal performance and can be extended with verification methods, demonstrating the potential of leveraging multiple LLMs in the generation-verification paradigm.
Referring video object segmentation (RVOS) is a challenging task that requires the model to segment the object in a video given the language description. MeViS is a recently proposed dataset that contains motion expressions of the target objects, leading to a challenging benchmark, compared with existing RVOS benchmarks. On the other hand, for referring expression tasks, a new trend is to adopt multi-modal large language model (MLLM) to achieve better image and text alignment. In this report, we show that with a simple modification to the test time inference method on stronger MLLMs, we can lead to stronger results on MeVIS. In particular, we adopt the recent method Sa2VA, a unified model for dense grounded understanding of both images and videos. By enlarging the scope of key frames, without any further training, we can achieve the 3rd place in the 4th PVUW workshop.
The future work section of a scientific article outlines potential research directions by identifying gaps and limitations of a current study. This section serves as a valuable resource for early-career researchers seeking unexplored areas and experienced researchers looking for new projects or collaborations. In this study, we generate future work suggestions from key sections of a scientific article alongside related papers and analyze how the trends have evolved. We experimented with various Large Language Models (LLMs) and integrated Retrieval-Augmented Generation (RAG) to enhance the generation process. We incorporate a LLM feedback mechanism to improve the quality of the generated content and propose an LLM-as-a-judge approach for evaluation. Our results demonstrated that the RAG-based approach with LLM feedback outperforms other methods evaluated through qualitative and quantitative metrics. Moreover, we conduct a human evaluation to assess the LLM as an extractor and judge. The code and dataset for this project are here, code: HuggingFace
Evaluating large language models (LLMs) in diverse and challenging scenarios is essential to align them with human preferences. To mitigate the prohibitive costs associated with human evaluations, utilizing a powerful LLM as a judge has emerged as a favored approach. Nevertheless, this methodology encounters several challenges, including substantial expenses, concerns regarding privacy and security, and reproducibility. In this paper, we propose a straightforward, replicable, and accurate automated evaluation method by leveraging a lightweight LLM as the judge, named RocketEval. Initially, we identify that the performance disparity between lightweight and powerful LLMs in evaluation tasks primarily stems from their ability to conduct comprehensive analyses, which is not easily enhanced through techniques such as chain-of-thought reasoning. By reframing the evaluation task as a multi-faceted Q&A using an instance-specific checklist, we demonstrate that the limited judgment accuracy of lightweight LLMs is largely attributes to high uncertainty and positional bias. To address these challenges, we introduce an automated evaluation process grounded in checklist grading, which is designed to accommodate a variety of scenarios and questions. This process encompasses the creation of checklists, the grading of these checklists by lightweight LLMs, and the reweighting of checklist items to align with the supervised annotations. Our experiments carried out on the automated evaluation benchmarks, MT-Bench and WildBench datasets, reveal that RocketEval, when using Gemma-2-2B as the judge, achieves a high correlation (0.965) with human preferences, which is comparable to GPT-4o. Moreover, RocketEval provides a cost reduction exceeding 50-fold for large-scale evaluation and comparison scenarios. Our code is available at https://github.com/Joinn99/RocketEval-ICLR .
The LLM-as-a-judge paradigm uses large language models (LLMs) for automated text evaluation, where a numerical assessment is assigned by an LLM to the input text following scoring rubrics. Existing methods for LLM-as-a-judge use cross-entropy (CE) loss for fine-tuning, which neglects the numeric nature of score prediction. Recent work addresses numerical prediction limitations of LLM fine-tuning through regression-aware fine-tuning, which, however, does not consider chain-of-thought (CoT) reasoning for score prediction. In this paper, we introduce TRACT (Two-stage Regression-Aware fine-tuning with CoT), a method combining CoT reasoning with regression-aware training. TRACT consists of two stages: first, seed LLM is fine-tuned to generate CoTs, which serve as supervision for the second stage fine-tuning. The training objective of TRACT combines the CE loss for learning the CoT reasoning capabilities, and the regression-aware loss for the score prediction. Experiments across four LLM-as-a-judge datasets and two LLMs show that TRACT significantly outperforms existing methods. Extensive ablation studies validate the importance of each component in TRACT.
Unsupervised panoptic segmentation aims to partition an image into semantically meaningful regions and distinct object instances without training on manually annotated data. In contrast to prior work on unsupervised panoptic scene understanding, we eliminate the need for object-centric training data, enabling the unsupervised understanding of complex scenes. To that end, we present the first unsupervised panoptic method that directly trains on scene-centric imagery. In particular, we propose an approach to obtain high-resolution panoptic pseudo labels on complex scene-centric data, combining visual representations, depth, and motion cues. Utilizing both pseudo-label training and a panoptic self-training strategy yields a novel approach that accurately predicts panoptic segmentation of complex scenes without requiring any human annotations. Our approach significantly improves panoptic quality, e.g., surpassing the recent state of the art in unsupervised panoptic segmentation on Cityscapes by 9.4% points in PQ.