llm - 2025_05
Navigation
- Part 1
- Part 2
- Part 3
- Part 4
- Part 5
- Part 6
- Part 7
- Part 8
- Part 9
- Part 10
- Part 11
- Part 12
- Part 13
- Part 14
- Part 15
- Part 16
- Part 17
Papers
Recently, multimodal large language models (MLLMs) have emerged as a key approach in achieving artificial general intelligence. In particular, vision-language MLLMs have been developed to generate not only text but also visual outputs from multimodal inputs. This advancement requires efficient image tokens that LLMs can process effectively both in input and output. However, existing image tokenization methods for MLLMs typically capture only global abstract concepts or uniformly segmented image patches, restricting MLLMs' capability to effectively understand or generate detailed visual content, particularly at the object level. To address this limitation, we propose an object-centric visual tokenizer based on Slot Attention specifically for MLLMs. In particular, based on the Q-Former encoder, diffusion decoder, and residual vector quantization, our proposed discretized slot tokens can encode local visual details while maintaining high-level semantics, and also align with textual data to be integrated seamlessly within a unified next-token prediction framework of LLMs. The resulting Slot-MLLM demonstrates significant performance improvements over baselines with previous visual tokenizers across various vision-language tasks that entail local detailed comprehension and generation. Notably, this work is the first demonstration of the feasibility of object-centric slot attention performed with MLLMs and in-the-wild natural images.
The remarkable reasoning and generalization capabilities of Large Language Models (LLMs) have paved the way for their expanding applications in embodied AI, robotics, and other real-world tasks. To effectively support these applications, grounding in spatial and temporal understanding in multimodal environments is essential. To this end, recent works have leveraged scene graphs, a structured representation that encodes entities, attributes, and their relationships in a scene. However, a comprehensive evaluation of LLMs' ability to utilize scene graphs remains limited. In this work, we introduce Text-Scene Graph (TSG) Bench, a benchmark designed to systematically assess LLMs' ability to (1) understand scene graphs and (2) generate them from textual narratives. With TSG Bench we evaluate 11 LLMs and reveal that, while models perform well on scene graph understanding, they struggle with scene graph generation, particularly for complex narratives. Our analysis indicates that these models fail to effectively decompose discrete scenes from a complex narrative, leading to a bottleneck when generating scene graphs. These findings underscore the need for improved methodologies in scene graph generation and provide valuable insights for future research. The demonstration of our benchmark is available at https://tsg-bench.netlify.app. Additionally, our code and evaluation data are publicly available at https://anonymous.4open.science/r/TSG-Bench.
Efficient multi-hop reasoning requires Large Language Models (LLMs) based agents to acquire high-value external knowledge iteratively. Previous work has explored reinforcement learning (RL) to train LLMs to perform search-based document retrieval, achieving notable improvements in QA performance, but underperform on complex, multi-hop QA resulting from the sparse rewards from global signal only. To address this gap in existing research, we introduce StepSearch, a framework for search LLMs that trained with step-wise proximal policy optimization method. It consists of richer and more detailed intermediate search rewards and token-level process supervision based on information gain and redundancy penalties to better guide each search step. We constructed a fine-grained question-answering dataset containing sub-question-level search trajectories based on open source datasets through a set of data pipeline method. On standard multi-hop QA benchmarks, it significantly outperforms global-reward baselines, achieving 11.2% and 4.2% absolute improvements for 3B and 7B models over various search with RL baselines using only 19k training data, demonstrating the effectiveness of fine-grained, stepwise supervision in optimizing deep search LLMs. Our code will be released on https://github.com/Zillwang/StepSearch.
Large Language Models (LLMs) represent substantial intellectual and economic investments, yet their effectiveness can inadvertently facilitate model imitation via knowledge distillation (KD).In practical scenarios, competitors can distill proprietary LLM capabilities by simply observing publicly accessible outputs, akin to reverse-engineering a complex performance by observation alone. Existing protective methods like watermarking only identify imitation post-hoc, while other defenses assume the student model mimics the teacher's internal logits, rendering them ineffective against distillation purely from observed output text. This paper confronts the challenge of actively protecting LLMs within the realistic constraints of API-based access. We introduce an effective and efficient Defensive Output Generation (DOGe) strategy that subtly modifies the output behavior of an LLM. Its outputs remain accurate and useful for legitimate users, yet are designed to be misleading for distillation, significantly undermining imitation attempts. We achieve this by fine-tuning only the final linear layer of the teacher LLM with an adversarial loss. This targeted training approach anticipates and disrupts distillation attempts during inference time. Our experiments show that, while preserving or even improving the original performance of the teacher model, student models distilled from the defensively generated teacher outputs demonstrate catastrophically reduced performance, demonstrating our method's effectiveness as a practical safeguard against KD-based model imitation.
Robots need task planning methods to achieve goals that require more than individual actions. Recently, large language models (LLMs) have demonstrated impressive performance in task planning. LLMs can generate a step-by-step solution using a description of actions and the goal. Despite the successes in LLM-based task planning, there is limited research studying the security aspects of those systems. In this paper, we develop Robo-Troj, the first multi-trigger backdoor attack for LLM-based task planners, which is the main contribution of this work. As a multi-trigger attack, Robo-Troj is trained to accommodate the diversity of robot application domains. For instance, one can use unique trigger words, e.g., "herical", to activate a specific malicious behavior, e.g., cutting hand on a kitchen robot. In addition, we develop an optimization method for selecting the trigger words that are most effective. Through demonstrating the vulnerability of LLM-based planners, we aim to promote the development of secured robot systems.
The Linux kernel is a critical system, serving as the foundation for numerous systems. Bugs in the Linux kernel can cause serious consequences, affecting billions of users. Fault localization (FL), which aims at identifying the buggy code elements in software, plays an essential role in software quality assurance. While recent LLM agents have achieved promising accuracy in FL on recent benchmarks like SWE-bench, it remains unclear how well these methods perform in the Linux kernel, where FL is much more challenging due to the large-scale code base, limited observability, and diverse impact factors. In this paper, we introduce LinuxFLBench, a FL benchmark constructed from real-world Linux kernel bugs. We conduct an empirical study to assess the performance of state-of-the-art LLM agents on the Linux kernel. Our initial results reveal that existing agents struggle with this task, achieving a best top-1 accuracy of only 41.6% at file level. To address this challenge, we propose LinuxFL$^+$, an enhancement framework designed to improve FL effectiveness of LLM agents for the Linux kernel. LinuxFL$^+$ substantially improves the FL accuracy of all studied agents (e.g., 7.2% - 11.2% accuracy increase) with minimal costs. Data and code are available at https://github.com/FudanSELab/LinuxFLBench.
CulFiT: A Fine-grained Cultural-aware LLM Training Paradigm via Multilingual Critique Data Synthesis
Large Language Models (LLMs) have demonstrated remarkable capabilities across various tasks, yet they often exhibit a specific cultural biases, neglecting the values and linguistic diversity of low-resource regions. This cultural bias not only undermines universal equality, but also risks reinforcing stereotypes and perpetuating discrimination. To address this, we propose CulFiT, a novel culturally-aware training paradigm that leverages multilingual data and fine-grained reward modeling to enhance cultural sensitivity and inclusivity. Our approach synthesizes diverse cultural-related questions, constructs critique data in culturally relevant languages, and employs fine-grained rewards to decompose cultural texts into verifiable knowledge units for interpretable evaluation. We also introduce GlobalCultureQA, a multilingual open-ended question-answering dataset designed to evaluate culturally-aware responses in a global context. Extensive experiments on three existing benchmarks and our GlobalCultureQA demonstrate that CulFiT achieves state-of-the-art open-source model performance in cultural alignment and general reasoning.
Large language models (LLMs) have shown remarkable performance across diverse reasoning and generation tasks, and are increasingly deployed as agents in dynamic environments such as code generation and recommendation systems. However, many real-world applications, such as high-frequency trading and real-time competitive gaming, require decisions under strict latency constraints, where faster responses directly translate into higher rewards. Despite the importance of this latency quality trade off, it remains underexplored in the context of LLM based agents. In this work, we present the first systematic study of this trade off in real time decision making tasks. To support our investigation, we introduce two new benchmarks: HFTBench, a high frequency trading simulation, and StreetFighter, a competitive gaming platform. Our analysis reveals that optimal latency quality balance varies by task, and that sacrificing quality for lower latency can significantly enhance downstream performance. To address this, we propose FPX, an adaptive framework that dynamically selects model size and quantization level based on real time demands. Our method achieves the best performance on both benchmarks, improving win rate by up to 80% in Street Fighter and boosting daily yield by up to 26.52% in trading, underscoring the need for latency aware evaluation and deployment strategies for LLM based agents. These results demonstrate the critical importance of latency aware evaluation and deployment strategies for real world LLM based agents. Our benchmarks are available at Latency Sensitive Benchmarks.
Recent studies have shown that vector representations of contextual embeddings learned by pre-trained large language models (LLMs) are effective in various downstream tasks in numerical domains. Despite their significant benefits, the tendency of LLMs to hallucinate in such domains can have severe consequences in applications such as energy, nature, finance, healthcare, retail and transportation, among others. To guarantee prediction reliability and accuracy in numerical domains, it is necessary to open the black-box and provide performance guarantees through explanation. However, there is little theoretical understanding of when pre-trained language models help solve numeric downstream tasks. This paper seeks to bridge this gap by understanding when the next-word prediction capability of LLMs can be adapted to numerical domains through a novel analysis based on the concept of isotropy in the contextual embedding space. Specifically, we consider a log-linear model for LLMs in which numeric data can be predicted from its context through a network with softmax in the output layer of LLMs (i.e., language model head in self-attention). We demonstrate that, in order to achieve state-of-the-art performance in numerical domains, the hidden representations of the LLM embeddings must possess a structure that accounts for the shift-invariance of the softmax function. By formulating a gradient structure of self-attention in pre-trained models, we show how the isotropic property of LLM embeddings in contextual embedding space preserves the underlying structure of representations, thereby resolving the shift-invariance problem and providing a performance guarantee. Experiments show that different characteristics of numeric data and model architecture could have different impacts on isotropy.
Despite the success of recommender systems in alleviating information overload, fairness issues have raised concerns in recent years, potentially leading to unequal treatment for certain user groups. While efforts have been made to improve recommendation fairness, they often assume that users' sensitive attributes are available during model training. However, collecting sensitive information can be difficult, especially on platforms that involve no personal information disclosure. Therefore, we aim to improve recommendation fairness without any access to sensitive attributes. However, this is a non-trivial task because uncovering latent sensitive patterns from complicated user behaviors without explicit sensitive attributes can be difficult. Consequently, suboptimal estimates of sensitive distributions can hinder the fairness training process. To address these challenges, leveraging the remarkable reasoning abilities of Large Language Models (LLMs), we propose a novel LLM-enhanced framework for Fair recommendation withOut Sensitive Attributes (LLMFOSA). A Multi-Persona Sensitive Information Inference module employs LLMs with distinct personas that mimic diverse human perceptions to infer and distill sensitive information. Furthermore, a Confusion-Aware Sensitive Representation Learning module incorporates inference results and rationales to develop robust sensitive representations, considering the mislabeling confusion and collective consensus among agents. The model is then optimized by a formulated mutual information objective. Extensive experiments on two public datasets validate the effectiveness of LLMFOSA in improving fairness.
Large language models (LLMs) have shown remarkable capabilities in commonsense reasoning; however, some variations in questions can trigger incorrect responses. Do these models truly understand commonsense knowledge, or just memorize expression patterns? To investigate this question, we present the first extensive robustness evaluation of LLMs in commonsense reasoning. We introduce HellaSwag-Pro, a large-scale bilingual benchmark consisting of 11,200 cases, by designing and compiling seven types of question variants. To construct this benchmark, we propose a two-stage method to develop Chinese HellaSwag, a finely annotated dataset comprising 12,000 instances across 56 categories. We conduct extensive experiments on 41 representative LLMs, revealing that these LLMs are far from robust in commonsense reasoning. Furthermore, this robustness varies depending on the language in which the LLM is tested. This work establishes a high-quality evaluation benchmark, with extensive experiments offering valuable insights to the community in commonsense reasoning for LLMs.
As large language models (LLMs) continue to advance, their deployment often involves fine-tuning to enhance performance on specific downstream tasks. However, this customization is sometimes accompanied by misleading claims about the origins, raising significant concerns about transparency and trust within the open-source community. Existing model verification techniques typically assess functional, representational, and weight similarities. However, these approaches often struggle against obfuscation techniques, such as permutations and scaling transformations. To address this limitation, we propose a novel detection method Origin-Tracer that rigorously determines whether a model has been fine-tuned from a specified base model. This method includes the ability to extract the LoRA rank utilized during the fine-tuning process, providing a more robust verification framework. This framework is the first to provide a formalized approach specifically aimed at pinpointing the sources of model fine-tuning. We empirically validated our method on thirty-one diverse open-source models under conditions that simulate real-world obfuscation scenarios. We empirically analyze the effectiveness of our framework and finally, discuss its limitations. The results demonstrate the effectiveness of our approach and indicate its potential to establish new benchmarks for model verification.
Incorporating collaborative information (CI) effectively is crucial for leveraging LLMs in recommendation tasks. Existing approaches often encode CI using soft tokens or abstract identifiers, which introduces a semantic misalignment with the LLM's natural language pretraining and hampers knowledge integration. To address this, we propose expressing CI directly in natural language to better align with LLMs' semantic space. We achieve this by retrieving a curated set of the most relevant user behaviors in natural language form. However, identifying informative CI is challenging due to the complexity of similarity and utility assessment. To tackle this, we introduce a Self-assessing COllaborative REtrieval framework (SCORE) following the retrieve-rerank paradigm. First, a Collaborative Retriever (CAR) is developed to consider both collaborative patterns and semantic similarity. Then, a Self-assessing Reranker (SARE) leverages LLMs' own reasoning to assess and prioritize retrieved behaviors. Finally, the selected behaviors are prepended to the LLM prompt as natural-language CI to guide recommendation. Extensive experiments on two public datasets validate the effectiveness of SCORE in improving LLM-based recommendation.
Large language models excel in general tasks, yet assessing their reliability in logic-heavy, precision-critical domains like finance, law, and healthcare remains challenging. To address this, we introduce BizFinBench, the first benchmark specifically designed to evaluate LLMs in real-world financial applications. BizFinBench consists of 6,781 well-annotated queries in Chinese, spanning five dimensions: numerical calculation, reasoning, information extraction, prediction recognition, and knowledge-based question answering, grouped into nine fine-grained categories. The benchmark includes both objective and subjective metrics. We also introduce IteraJudge, a novel LLM evaluation method that reduces bias when LLMs serve as evaluators in objective metrics. We benchmark 25 models, including both proprietary and open-source systems. Extensive experiments show that no model dominates across all tasks. Our evaluation reveals distinct capability patterns: (1) In Numerical Calculation, Claude-3.5-Sonnet (63.18) and DeepSeek-R1 (64.04) lead, while smaller models like Qwen2.5-VL-3B (15.92) lag significantly; (2) In Reasoning, proprietary models dominate (ChatGPT-o3: 83.58, Gemini-2.0-Flash: 81.15), with open-source models trailing by up to 19.49 points; (3) In Information Extraction, the performance spread is the largest, with DeepSeek-R1 scoring 71.46, while Qwen3-1.7B scores 11.23; (4) In Prediction Recognition, performance variance is minimal, with top models scoring between 39.16 and 50.00. We find that while current LLMs handle routine finance queries competently, they struggle with complex scenarios requiring cross-concept reasoning. BizFinBench offers a rigorous, business-aligned benchmark for future research. The code and dataset are available at https://github.com/HiThink-Research/BizFinBench.
Large Language Models (LLMs) falter in multi-step interactions -- often hallucinating, repeating actions, or misinterpreting user corrections -- due to reliance on linear, unstructured context. This fragility stems from the lack of persistent memory to track evolving goals and task dependencies, undermining trust in autonomous agents. We introduce the Task Memory Engine (TME), a modular memory controller that transforms existing LLMs into robust, revision-aware agents without fine-tuning. TME implements a spatial memory framework that replaces flat context with graph-based structures to support consistent, multi-turn reasoning. Departing from linear concatenation and ReAct-style prompting, TME builds a dynamic task graph -- either a tree or directed acyclic graph (DAG) -- to map user inputs to subtasks, align them with prior context, and enable dependency-tracked revisions. Its Task Representation and Intent Management (TRIM) component models task semantics and user intent to ensure accurate interpretation. Across four multi-turn scenarios-trip planning, cooking, meeting scheduling, and shopping cart editing -- TME eliminates 100% of hallucinations and misinterpretations in three tasks, and reduces hallucinations by 66.7% and misinterpretations by 83.3% across 27 user turns, outperforming ReAct. TME's modular design supports plug-and-play deployment and domain-specific customization, adaptable to both personal assistants and enterprise automation. We release TME's codebase, benchmarks, and components as open-source resources, enabling researchers to develop reliable LLM agents. TME's scalable architecture addresses a critical gap in agent performance across complex, interactive settings.
We introduce a novel framework for consolidating multi-turn adversarial ``jailbreak'' prompts into single-turn queries, significantly reducing the manual overhead required for adversarial testing of large language models (LLMs). While multi-turn human jailbreaks have been shown to yield high attack success rates, they demand considerable human effort and time. Our multi-turn-to-single-turn (M2S) methods -- Hyphenize, Numberize, and Pythonize -- systematically reformat multi-turn dialogues into structured single-turn prompts. Despite removing iterative back-and-forth interactions, these prompts preserve and often enhance adversarial potency: in extensive evaluations on the Multi-turn Human Jailbreak (MHJ) dataset, M2S methods achieve attack success rates from 70.6 percent to 95.9 percent across several state-of-the-art LLMs. Remarkably, the single-turn prompts outperform the original multi-turn attacks by as much as 17.5 percentage points while cutting token usage by more than half on average. Further analysis shows that embedding malicious requests in enumerated or code-like structures exploits ``contextual blindness'', bypassing both native guardrails and external input-output filters. By converting multi-turn conversations into concise single-turn prompts, the M2S framework provides a scalable tool for large-scale red teaming and reveals critical weaknesses in contemporary LLM defenses.
Post-training compression reduces the computational and memory costs of large language models (LLMs), enabling resource-efficient deployment. However, existing compression benchmarks only focus on language modeling (e.g., perplexity) and natural language understanding tasks (e.g., GLUE accuracy), ignoring the agentic capabilities - workflow, tool use/function call, long-context understanding and real-world application. We introduce the Agent Compression Benchmark (ACBench), the first comprehensive benchmark for evaluating how compression impacts LLMs' agentic abilities. ACBench spans (1) 12 tasks across 4 capabilities (e.g., WorfBench for workflow generation, Needle-in-Haystack for long-context retrieval), (2) quantization (GPTQ, AWQ) and pruning (Wanda, SparseGPT), and (3) 15 models, including small (Gemma-2B), standard (Qwen2.5 7B-32B), and distilled reasoning LLMs (DeepSeek-R1-Distill). Our experiments reveal compression tradeoffs: 4-bit quantization preserves workflow generation and tool use (1%-3% drop) but degrades real-world application accuracy by 10%-15%. We introduce ERank, Top-k Ranking Correlation and Energy to systematize analysis. ACBench provides actionable insights for optimizing LLM compression in agentic scenarios. The code can be found in https://github.com/pprp/ACBench.
Recent efforts leverage Large Language Models (LLMs) for modeling text-attributed graph structures in node classification tasks. These approaches describe graph structures for LLMs to understand or aggregate LLM-generated textual attribute embeddings through graph structure. However, these approaches face two main limitations in modeling graph structures with LLMs. (i) Graph descriptions become verbose in describing high-order graph structure. (ii) Textual attributes alone do not contain adequate graph structure information. It is challenging to model graph structure concisely and adequately with LLMs. LLMs lack built-in mechanisms to model graph structures directly. They also struggle with complex long-range dependencies between high-order nodes and target nodes. Inspired by the observation that LLMs pre-trained on one language can achieve exceptional performance on another with minimal additional training, we propose \textbf{G}raph-\textbf{D}efined \textbf{L}anguage for \textbf{L}arge \textbf{L}anguage \textbf{M}odel (GDL4LLM). This novel framework enables LLMs to transfer their powerful language understanding capabilities to graph-structured data. GDL4LLM translates graphs into a graph language corpus instead of graph descriptions and pre-trains LLMs on this corpus to adequately understand graph structures. During fine-tuning, this corpus describes the structural information of target nodes concisely with only a few tokens. By treating graphs as a new language, GDL4LLM enables LLMs to model graph structures adequately and concisely for node classification tasks. Extensive experiments on three real-world datasets demonstrate that GDL4LLM outperforms description-based and textual attribute embeddings-based baselines by efficiently modeling different orders of graph structure with LLMs.
The quality of training data is critical to the performance of machine learning applications in domains like transportation, healthcare, and robotics. Accurate image labeling, however, often relies on time-consuming, expert-driven methods with limited feedback. This research introduces a sketch-based annotation approach supported by large language models (LLMs) to reduce technical barriers and enhance accessibility. Using a synthetic dataset, we examine how sketch recognition features relate to LLM feedback metrics, aiming to improve the reliability and interpretability of LLM-assisted labeling. We also explore how prompting strategies and sketch variations influence feedback quality. Our main contribution is a sketch-based virtual assistant that simplifies annotation for non-experts and advances LLM-driven labeling tools in terms of scalability, accessibility, and explainability.
Recently, large language models (LLMs) have demonstrated remarkable capabilities in natural language processing tasks, yet they remain prone to hallucinations when reasoning with insufficient internal knowledge. While integrating LLMs with knowledge graphs (KGs) provides access to structured, verifiable information, existing approaches often generate incomplete or factually inconsistent reasoning paths. To this end, we propose Self-Reflective Planning (SRP), a framework that synergizes LLMs with KGs through iterative, reference-guided reasoning. Specifically, given a question and topic entities, SRP first searches for references to guide planning and reflection. In the planning process, it checks initial relations and generates a reasoning path. After retrieving knowledge from KGs through a reasoning path, it implements iterative reflection by judging the retrieval result and editing the reasoning path until the answer is correctly retrieved. Extensive experiments on three public datasets demonstrate that SRP surpasses various strong baselines and further underscore its reliable reasoning ability.
As large language models (LLMs) evolve into autonomous agents capable of collaborative reasoning and task execution, multi-agent LLM systems have emerged as a powerful paradigm for solving complex problems. However, these systems pose new challenges for copyright protection, particularly when sensitive or copyrighted content is inadvertently recalled through inter-agent communication and reasoning. Existing protection techniques primarily focus on detecting content in final outputs, overlooking the richer, more revealing reasoning processes within the agents themselves. In this paper, we introduce CoTGuard, a novel framework for copyright protection that leverages trigger-based detection within Chain-of-Thought (CoT) reasoning. Specifically, we can activate specific CoT segments and monitor intermediate reasoning steps for unauthorized content reproduction by embedding specific trigger queries into agent prompts. This approach enables fine-grained, interpretable detection of copyright violations in collaborative agent scenarios. We evaluate CoTGuard on various benchmarks in extensive experiments and show that it effectively uncovers content leakage with minimal interference to task performance. Our findings suggest that reasoning-level monitoring offers a promising direction for safeguarding intellectual property in LLM-based agent systems.
The surge of LLM studies makes synthesizing their findings challenging. Analysis of experimental results from literature can uncover important trends across studies, but the time-consuming nature of manual data extraction limits its use. Our study presents a semi-automated approach for literature analysis that accelerates data extraction using LLMs. It automatically identifies relevant arXiv papers, extracts experimental results and related attributes, and organizes them into a structured dataset, LLMEvalDB. We then conduct an automated literature analysis of frontier LLMs, reducing the effort of paper surveying and data extraction by more than 93% compared to manual approaches. We validate LLMEvalDB by showing that it reproduces key findings from a recent manual analysis of Chain-of-Thought (CoT) reasoning and also uncovers new insights that go beyond it, showing, for example, that in-context examples benefit coding & multimodal tasks but offer limited gains in math reasoning tasks compared to zero-shot CoT. Our automatically updatable dataset enables continuous tracking of target models by extracting evaluation studies as new data becomes available. Through LLMEvalDB and empirical analysis, we provide insights into LLMs while facilitating ongoing literature analyses of their behavior.
Incorporating external knowledge has emerged as a promising way to mitigate outdated knowledge and hallucinations in LLM. However, external knowledge is often imperfect, encompassing substantial extraneous or even inaccurate content, which interferes with the LLM's utilization of useful knowledge in the context. This paper seeks to characterize the features of preferred external knowledge and perform empirical studies in imperfect contexts. Inspired by the chain of evidence (CoE), we characterize that the knowledge preferred by LLMs should maintain both relevance to the question and mutual support among the textual pieces. Accordingly, we propose a CoE discrimination approach and conduct a comparative analysis between CoE and Non-CoE samples across significance, deceptiveness, and robustness, revealing the LLM's preference for external knowledge that aligns with CoE features. Furthermore, we selected three representative tasks (RAG-based multi-hop QA, external knowledge poisoning and poisoning defense), along with corresponding SOTA or prevalent baselines. By integrating CoE features, the variants achieved significant improvements over the original baselines.
In this paper, we conduct a critical review of existing theories and frameworks on human-human collaborative writing to assess their relevance to the current human-AI paradigm in professional contexts, and draw seven insights along with design implications for human-AI collaborative writing tools. We found that, as LLMs nudge the writing process more towards an empirical "trial and error" process analogous to prototyping, the non-linear cognitive process of writing will stay the same, but more rigor will be required for revision methodologies. This shift would shed further light on the importance of coherence support, but the large language model (LLM)'s unprecedented semantic capabilities can bring novel approaches to this ongoing challenge. We argue that teamwork-related factors such as group awareness, consensus building and authorship - which have been central in human-human collaborative writing studies - should not apply to the human-AI paradigm due to excessive anthropomorphism. With the LLM's text generation capabilities becoming essentially indistinguishable from human-written ones, we are entering an era where, for the first time in the history of computing, we are engaging in collaborative writing with AI at workplaces on a daily basis. We aim to bring theoretical grounding and practical design guidance to the interaction designs of human-AI collaborative writing, with the goal of enhancing future human-AI writing software.
Arabic poetry is one of the richest and most culturally rooted forms of expression in the Arabic language, known for its layered meanings, stylistic diversity, and deep historical continuity. Although large language models (LLMs) have demonstrated strong performance across languages and tasks, their ability to understand Arabic poetry remains largely unexplored. In this work, we introduce \emph{Fann or Flop}, the first benchmark designed to assess the comprehension of Arabic poetry by LLMs in 12 historical eras, covering 14 core poetic genres and a variety of metrical forms, from classical structures to contemporary free verse. The benchmark comprises a curated corpus of poems with explanations that assess semantic understanding, metaphor interpretation, prosodic awareness, and cultural context. We argue that poetic comprehension offers a strong indicator for testing how good the LLM understands classical Arabic through Arabic poetry. Unlike surface-level tasks, this domain demands deeper interpretive reasoning and cultural sensitivity. Our evaluation of state-of-the-art LLMs shows that most models struggle with poetic understanding despite strong results on standard Arabic benchmarks. We release "Fann or Flop" along with the evaluation suite as an open-source resource to enable rigorous evaluation and advancement for Arabic language models. Code is available at: https://github.com/mbzuai-oryx/FannOrFlop.
Large language models (LLMs) have demonstrated strong performance in various robot control tasks. However, their deployment in real-world applications remains constrained. Even state-ofthe-art LLMs, such as GPT-o4mini, frequently produce invalid action plans that violate physical constraints, such as directing a robot to an unreachable location or causing collisions between robots. This issue primarily arises from a lack of awareness of these physical constraints during the reasoning process. To address this issue, we propose a novel framework that integrates reinforcement learning with verifiable rewards (RLVR) to incentivize knowledge of physical constraints into LLMs to induce constraints-aware reasoning during plan generation. In this approach, only valid action plans that successfully complete a control task receive positive rewards. We applied our method to two small-scale LLMs: a non-reasoning Qwen2.5-3B-Instruct and a reasoning Qwen3-4B. The experiment results demonstrate that constraint-aware small LLMs largely outperform large-scale models without constraints, grounded on both the BoxNet task and a newly developed BoxNet3D environment built using MuJoCo. This work highlights the effectiveness of grounding even small LLMs with physical constraints to enable scalable and efficient multi-robot control in complex, physically constrained environments.
Large Language Models (LLMs) trained via Reinforcement Learning (RL) have exhibited strong reasoning capabilities and emergent reflective behaviors, such as backtracking and error correction. However, conventional Markovian RL confines exploration to the training phase to learn an optimal deterministic policy and depends on the history contexts only through the current state. Therefore, it remains unclear whether reflective reasoning will emerge during Markovian RL training, or why they are beneficial at test time. To remedy this, we recast reflective exploration within the Bayes-Adaptive RL framework, which explicitly optimizes the expected return under a posterior distribution over Markov decision processes. This Bayesian formulation inherently incentivizes both reward-maximizing exploitation and information-gathering exploration via belief updates. Our resulting algorithm, BARL, instructs the LLM to stitch and switch strategies based on the observed outcomes, offering principled guidance on when and how the model should reflectively explore. Empirical results on both synthetic and mathematical reasoning tasks demonstrate that BARL outperforms standard Markovian RL approaches at test time, achieving superior token efficiency with improved exploration effectiveness. Our code is available at https://github.com/shenao-zhang/BARL.
Reinforcement learning-based post-training of large language models (LLMs) has recently gained attention, particularly following the release of DeepSeek R1, which applied GRPO for fine-tuning. Amid the growing hype around improved reasoning abilities attributed to RL post-training, we critically examine the formulation and assumptions underlying these methods. We start by highlighting the popular structural assumptions made in modeling LLM training as a Markov Decision Process (MDP), and show how they lead to a degenerate MDP that doesn't quite need the RL/GRPO apparatus. The two critical structural assumptions include (1) making the MDP states be just a concatenation of the actions-with states becoming the context window and the actions becoming the tokens in LLMs and (2) splitting the reward of a state-action trajectory uniformly across the trajectory. Through a comprehensive analysis, we demonstrate that these simplifying assumptions make the approach effectively equivalent to an outcome-driven supervised learning. Our experiments on benchmarks including GSM8K and Countdown using Qwen-2.5 base models show that iterative supervised fine-tuning, incorporating both positive and negative samples, achieves performance comparable to GRPO-based training. We will also argue that the structural assumptions indirectly incentivize the RL to generate longer sequences of intermediate tokens-which in turn feeds into the narrative of "RL generating longer thinking traces." While RL may well be a very useful technique for improving the reasoning abilities of LLMs, our analysis shows that the simplistic structural assumptions made in modeling the underlying MDP render the popular LLM RL frameworks and their interpretations questionable.
Humans organize knowledge into compact categories through semantic compression by mapping diverse instances to abstract representations while preserving meaning (e.g., robin and blue jay are both birds; most birds can fly). These concepts reflect a trade-off between expressive fidelity and representational simplicity. Large Language Models (LLMs) demonstrate remarkable linguistic abilities, yet whether their internal representations strike a human-like trade-off between compression and semantic fidelity is unclear. We introduce a novel information-theoretic framework, drawing from Rate-Distortion Theory and the Information Bottleneck principle, to quantitatively compare these strategies. Analyzing token embeddings from a diverse suite of LLMs against seminal human categorization benchmarks, we uncover key divergences. While LLMs form broad conceptual categories that align with human judgment, they struggle to capture the fine-grained semantic distinctions crucial for human understanding. More fundamentally, LLMs demonstrate a strong bias towards aggressive statistical compression, whereas human conceptual systems appear to prioritize adaptive nuance and contextual richness, even if this results in lower compressional efficiency by our measures. These findings illuminate critical differences between current AI and human cognitive architectures, guiding pathways toward LLMs with more human-aligned conceptual representations.
Despite the significant potential of FP8 data formats for large language model (LLM) pre-training, their adoption has been limited due to challenges in maintaining stability at scale. Existing approaches often rely on suboptimal fine-grained FP8 kernels or fall back to higher-precision matrix multiplications (GEMMs) in sensitive components, such as attention projections, compromising potential throughput gains. We introduce a new class of LLM architectures that, for the first time, support FP8 computation for all GEMMs within transformer blocks during both forward and backward passes. This enables unprecedented throughput gains, particularly at scale, while matching the downstream performance of standard BF16 training. Our architecture design reduces large outlier activations, promoting stable long-term FP8 training. In addition, we identify key metrics to monitor low-precision training and predict potential future divergences.
This paper presents Project Riley, a novel multimodal and multi-model conversational AI architecture oriented towards the simulation of reasoning influenced by emotional states. Drawing inspiration from Pixar's Inside Out, the system comprises five distinct emotional agents - Joy, Sadness, Fear, Anger, and Disgust - that engage in structured multi-round dialogues to generate, criticise, and iteratively refine responses. A final reasoning mechanism synthesises the contributions of these agents into a coherent output that either reflects the dominant emotion or integrates multiple perspectives. The architecture incorporates both textual and visual large language models (LLMs), alongside advanced reasoning and self-refinement processes. A functional prototype was deployed locally in an offline environment, optimised for emotional expressiveness and computational efficiency. From this initial prototype, another one emerged, called Armando, which was developed for use in emergency contexts, delivering emotionally calibrated and factually accurate information through the integration of Retrieval-Augmented Generation (RAG) and cumulative context tracking. The Project Riley prototype was evaluated through user testing, in which participants interacted with the chatbot and completed a structured questionnaire assessing three dimensions: Emotional Appropriateness, Clarity and Utility, and Naturalness and Human-likeness. The results indicate strong performance in structured scenarios, particularly with respect to emotional alignment and communicative clarity.
Lipograms are a unique form of constrained writing where all occurrences of a particular letter are excluded from the text, typified by the novel Gadsby, which daringly avoids all usage of the letter 'e'. In this study, we explore the power of modern large language models (LLMs) by transforming the novel F. Scott Fitzgerald's The Great Gatsby into a fully 'e'-less text. We experimented with a range of techniques, from baseline methods like synonym replacement to sophisticated generative models enhanced with beam search and named entity analysis. We show that excluding up to 3.6% of the most common letters (up to the letter 'u') had minimal impact on the text's meaning, although translation fidelity rapidly and predictably decays with stronger lipogram constraints. Our work highlights the surprising flexibility of English under strict constraints, revealing just how adaptable and creative language can be.
Factual completeness is a general term that captures how detailed and informative a factually correct text is. For instance, the factual sentence ``Barack Obama was born in the United States'' is factually correct, though less informative than the factual sentence ``Barack Obama was born in Honolulu, Hawaii, United States''. Despite the known fact that LLMs tend to hallucinate and generate factually incorrect text, they might also tend to choose to generate factual text that is indeed factually correct and yet less informative than other, more informative choices. In this work, we tackle this problem by proposing an informativeness alignment mechanism. This mechanism takes advantage of recent factual benchmarks to propose an informativeness alignment objective. This objective prioritizes answers that are both correct and informative. A key finding of our work is that when training a model to maximize this objective or optimize its preference, we can improve not just informativeness but also factuality.
Today, large language models are widely used as judges to evaluate responses from other language models. Hence, it is imperative to benchmark and improve these LLM-judges on real-world language model usage: a typical human-assistant conversation is lengthy, and shows significant diversity in topics, intents, and requirements across turns, e.g. social interactions, task requests, feedback. We present Amulet, a framework that leverages pertinent linguistic concepts of dialog-acts and maxims to improve the accuracy of LLM-judges on preference data with complex, multi-turn conversational context. Amulet presents valuable insights about (a) the communicative structures and intents present in the conversation (dialog acts), and (b) the satisfaction of conversational principles (maxims) by the preference responses, and uses them to make judgments. On four challenging datasets, Amulet shows that (a) humans frequently (60 to 70 percent of the time) change their intents from one turn of the conversation to the next, and (b) in 75 percent of instances, the preference responses can be differentiated via dialog acts and/or maxims, reiterating the latter's significance in judging such data. Amulet can be used either as a judge by applying the framework to a single LLM, or integrated into a jury with different LLM judges; our judges and juries show strong improvements on relevant baselines for all four datasets.
The growing demand for efficient Large Language Model (LLM) inference requires a holistic optimization on algorithms, systems, and hardware. However, very few works have fundamentally changed the generation pattern: each token needs one forward pass and one KV cache. This can be sub-optimal because we found that LLMs are extremely capable of self-identifying the exact dose of information that a single KV cache can store, and many tokens can be generated confidently without global context. Based on this insight, we introduce HAMburger, a Hierarchically Auto-regressive Model that redefines resource allocation in LLMs by moving beyond uniform computation and storage per token during inference. Stacking a compositional embedder and a micro-step decoder in between a base LLM, HAMburger smashes multiple tokens into a single KV and generates several tokens per step. Additionally, HAMburger functions as a speculative decoding framework where it can blindly trust self-drafted tokens. As a result, HAMburger shifts the growth of KV cache and forward FLOPs from linear to sub-linear with respect to output length, and adjusts its inference speed based on query perplexity and output structure. Extensive evaluations show that HAMburger reduces the KV cache computation by up to 2$\times$ and achieves up to 2$\times$ TPS, while maintaining quality in both short- and long-context tasks. Our method explores an extremely challenging inference regime that requires both computation- and memory-efficiency with a hardware-agnostic design.
Fine-tuning for large language models (LLMs) typically requires substantial amounts of high-quality supervised data, which is both costly and labor-intensive to acquire. While synthetic data generation has emerged as a promising solution, existing approaches frequently suffer from factual inaccuracies, insufficient long-tail coverage, simplistic knowledge structures, and homogenized outputs. To address these challenges, we introduce GraphGen, a knowledge graph-guided framework designed for three key question-answering (QA) scenarios: atomic QA, aggregated QA, and multi-hop QA. It begins by constructing a fine-grained knowledge graph from the source text. It then identifies knowledge gaps in LLMs using the expected calibration error metric, prioritizing the generation of QA pairs that target high-value, long-tail knowledge. Furthermore, GraphGen incorporates multi-hop neighborhood sampling to capture complex relational information and employs style-controlled generation to diversify the resulting QA data. Experimental results on knowledge-intensive tasks under closed-book settings demonstrate that GraphGen outperforms conventional synthetic data methods, offering a more reliable and comprehensive solution to the data scarcity challenge in supervised fine-tuning. The code and data are publicly available at https://github.com/open-sciencelab/GraphGen.
To reveal when a large language model (LLM) is uncertain about a response, uncertainty quantification commonly produces percentage numbers along with the output. But is this all we can do? We argue that in the output space of LLMs, the space of strings, exist strings expressive enough to summarize the distribution over output strings the LLM deems possible. We lay a foundation for this new avenue of uncertainty explication and present SelfReflect, a theoretically-motivated metric to assess how faithfully a string summarizes an LLM's internal answer distribution. We show that SelfReflect is able to discriminate even subtle differences of candidate summary strings and that it aligns with human judgement, outperforming alternative metrics such as LLM judges and embedding comparisons. With SelfReflect, we investigate a number of self-summarization methods and find that even state-of-the-art reasoning models struggle to explicate their internal uncertainty. But we find that faithful summarizations can be generated by sampling and summarizing. Our metric enables future works towards this universal form of LLM uncertainties.
Large Language Models (LLMs) have demonstrated impressive reasoning abilities through test-time computation (TTC) techniques such as chain-of-thought prompting and tree-based reasoning. However, we argue that current reasoning LLMs (RLLMs) lack the ability to systematically explore the solution space. This paper formalizes what constitutes systematic problem solving and identifies common failure modes that reveal reasoning LLMs to be wanderers rather than systematic explorers. Through qualitative and quantitative analysis across multiple state-of-the-art LLMs, we uncover persistent issues: invalid reasoning steps, redundant explorations, hallucinated or unfaithful conclusions, and so on. Our findings suggest that current models' performance can appear to be competent on simple tasks yet degrade sharply as complexity increases. Based on the findings, we advocate for new metrics and tools that evaluate not just final outputs but the structure of the reasoning process itself.
Training large language models (LLMs) as interactive agents presents unique challenges including long-horizon decision making and interacting with stochastic environment feedback. While reinforcement learning (RL) has enabled progress in static tasks, multi-turn agent RL training remains underexplored. We propose StarPO (State-Thinking-Actions-Reward Policy Optimization), a general framework for trajectory-level agent RL, and introduce RAGEN, a modular system for training and evaluating LLM agents. Our study on four stylized environments reveals three core findings. First, our agent RL training shows a recurring mode of Echo Trap where reward variance cliffs and gradient spikes; we address this with StarPO-S, a stabilized variant with trajectory filtering, critic incorporation, and gradient stabilization. Second, we find the shaping of RL rollouts would benefit from diverse initial states, medium interaction granularity and more frequent sampling. Third, we show that without fine-grained, reasoning-aware reward signals, agent reasoning hardly emerge through multi-turn RL and they may show shallow strategies or hallucinated thoughts. Code and environments are available at https://github.com/RAGEN-AI/RAGEN.
Limited access to mental healthcare, extended wait times, and increasing capabilities of Large Language Models (LLMs) has led individuals to turn to LLMs for fulfilling their mental health needs. However, examining the multi-turn mental health conversation capabilities of LLMs remains under-explored. Existing evaluation frameworks typically focus on diagnostic accuracy and win-rates and often overlook alignment with patient-specific goals, values, and personalities required for meaningful conversations. To address this, we introduce MedAgent, a novel framework for synthetically generating realistic, multi-turn mental health sensemaking conversations and use it to create the Mental Health Sensemaking Dialogue (MHSD) dataset, comprising over 2,200 patient-LLM conversations. Additionally, we present MultiSenseEval, a holistic framework to evaluate the multi-turn conversation abilities of LLMs in healthcare settings using human-centric criteria. Our findings reveal that frontier reasoning models yield below-par performance for patient-centric communication and struggle at advanced diagnostic capabilities with average score of 31%. Additionally, we observed variation in model performance based on patient's persona and performance drop with increasing turns in the conversation. Our work provides a comprehensive synthetic data generation framework, a dataset and evaluation framework for assessing LLMs in multi-turn mental health conversations.
Fine-tuning large language models (LLMs) is intended to improve their reasoning capabilities, yet we uncover a counterintuitive effect: models often forget how to solve problems they previously answered correctly during training. We term this phenomenon temporal forgetting and show that it is widespread across model sizes, fine-tuning methods (both Reinforcement Learning and Supervised Fine-Tuning), and multiple reasoning benchmarks. To address this gap, we introduce Temporal Sampling, a simple decoding strategy that draws outputs from multiple checkpoints along the training trajectory. This approach recovers forgotten solutions without retraining or ensembling, and leads to substantial improvements in reasoning performance, gains from 4 to 19 points in Pass@k and consistent gains in Majority@k across several benchmarks. We further extend our method to LoRA-adapted models, demonstrating that storing only adapter weights across checkpoints achieves similar benefits with minimal storage cost. By leveraging the temporal diversity inherent in training, Temporal Sampling offers a practical, compute-efficient way to surface hidden reasoning ability and rethink how we evaluate LLMs.
The advanced role-playing capabilities of Large Language Models (LLMs) have enabled rich interactive scenarios, yet existing research in social interactions neglects hallucination while struggling with poor generalizability and implicit character fidelity judgments. To bridge this gap, motivated by human behaviour, we introduce a generalizable and explicit paradigm for uncovering interactive patterns of LLMs across diverse worldviews. Specifically, we first define interactive hallucination through stance transfer, then construct SHARP, a benchmark built by extracting relations from commonsense knowledge graphs and utilizing LLMs' inherent hallucination properties to simulate multi-role interactions. Extensive experiments confirm our paradigm's effectiveness and stability, examine the factors that influence these metrics, and challenge conventional hallucination mitigation solutions. More broadly, our work reveals a fundamental limitation in popular post-training methods for role-playing LLMs: the tendency to obscure knowledge beneath style, resulting in monotonous yet human-like behaviors - interactive hallucination.
Serving systems for Large Language Models (LLMs) are often optimized to improve quality of service (QoS) and throughput. However, due to the lack of open-source LLM serving workloads, these systems are frequently evaluated under unrealistic workload assumptions. Consequently, performance may degrade when systems are deployed in real-world scenarios. This work presents BurstGPT, an LLM serving workload with 10.31 million traces from regional Azure OpenAI GPT services over 213 days. BurstGPT captures LLM serving characteristics from user, model and system perspectives: (1) User request concurrency: burstiness variations of requests in Azure OpenAI GPT services, revealing diversified concurrency patterns in different services and model types. (2) User conversation patterns: counts and intervals within conversations for service optimizations. (3) Model response lengths: auto-regressive serving processes of GPT models, showing statistical relations between requests and their responses. (4) System response failures: failures of conversation and API services, showing intensive resource needs and limited availability of LLM services in Azure. The details of the characteristics can serve multiple purposes in LLM serving optimizations, such as system evaluation and trace provisioning. In our demo evaluation with BurstGPT, frequent variations in BurstGPT reveal declines in efficiency, stability, or reliability in realistic LLM serving. We identify that the generalization of KV cache management, scheduling and disaggregation optimizations can be improved under realistic workload evaluations. BurstGPT is publicly available now at https://github.com/HPMLL/BurstGPT and is widely used to develop prototypes of LLM serving frameworks in the industry.
As large language models grow in capability and agency, identifying vulnerabilities through red-teaming becomes vital for safe deployment. However, traditional prompt-engineering approaches may prove ineffective once red-teaming turns into a weak-to-strong problem, where target models surpass red-teamers in capabilities. To study this shift, we frame red-teaming through the lens of the capability gap between attacker and target. We evaluate more than 500 attacker-target pairs using LLM-based jailbreak attacks that mimic human red-teamers across diverse families, sizes, and capability levels. Three strong trends emerge: (i) more capable models are better attackers, (ii) attack success drops sharply once the target's capability exceeds the attacker's, and (iii) attack success rates correlate with high performance on social science splits of the MMLU-Pro benchmark. From these trends, we derive a jailbreaking scaling law that predicts attack success for a fixed target based on attacker-target capability gap. These findings suggest that fixed-capability attackers (e.g., humans) may become ineffective against future models, increasingly capable open-source models amplify risks for existing systems, and model providers must accurately measure and control models' persuasive and manipulative abilities to limit their effectiveness as attackers.
Effective generalization in language models depends critically on the diversity of their training data. Yet existing diversity metrics often fall short of this goal, relying on surface-level heuristics that are decoupled from model behavior. This motivates us to ask: What kind of diversity in training data actually drives generalization in language models -- and how can we measure and amplify it? Through large-scale empirical analyses spanning over 300 training runs, carefully controlled for data scale and quality, we show that data diversity can be a strong predictor of generalization in LLM reasoning -- as measured by average model performance on unseen out-of-distribution benchmarks. We introduce G-Vendi, a metric that quantifies diversity via the entropy of model-induced gradients. Despite using a small off-the-shelf proxy model for gradients, G-Vendi consistently outperforms alternative measures, achieving strong correlation (Spearman's $\rho \approx 0.9$) with out-of-distribution (OOD) performance on both natural language inference (NLI) and math reasoning tasks. Building on this insight, we present Prismatic Synthesis, a framework for generating diverse synthetic data by targeting underrepresented regions in gradient space. Experimental results show that Prismatic Synthesis consistently improves model performance as we scale synthetic data -- not just on in-distribution test but across unseen, out-of-distribution benchmarks -- significantly outperforming state-of-the-art models that rely on 20 times larger data generator than ours. For example, PrismMath-7B, our model distilled from a 32B LLM, outperforms R1-Distill-Qwen-7B -- the same base model trained on proprietary data generated by 671B R1 -- on 6 out of 7 challenging benchmarks.
Binary decompilation plays a vital role in various cybersecurity and software engineering tasks. Recently, end-to-end decompilation methods powered by large language models (LLMs) have garnered significant attention due to their ability to generate highly readable source code with minimal human intervention. However, existing LLM-based approaches face several critical challenges, including limited capability in reconstructing code structure and logic, low accuracy in data recovery, concerns over data security and privacy, and high computational resource requirements. To address these issues, we develop the CodeInverter Suite, making three contributions: (1) the CodeInverter Workflow (CIW) is a novel prompt engineering workflow that incorporates control flow graphs (CFG) and explicit data mappings to improve LLM-based decompilation. (2) Using CIW on well-known source code datasets, we curate the CodeInverter Dataset (CID), a domain-specific dataset containing 8.69 million samples that contains CFGs and data mapping tables. (3) We train the CoderInverter Models (CIMs) on CID, generating two lightweight LLMs (with 1.3B and 6.7B parameters) intended for efficient inference in privacy-sensitive or resource-constrained environments. Extensive experiments on two benchmarks demonstrate that the CIW substantially enhances the performance of various LLMs across multiple metrics. Our CIM-6.7B can achieve state-of-the-art decompilation performance, outperforming existing LLMs even with over 100x more parameters in decompilation tasks, an average improvement of 11.03% in re-executability, 6.27% in edit similarity.
Electronic Health Records (EHRs) are digital records of patient information, often containing unstructured clinical text. Named Entity Recognition (NER) is essential in EHRs for extracting key medical entities like problems, tests, and treatments to support downstream clinical applications. This paper explores prompt-based medical entity recognition using large language models (LLMs), specifically GPT-4o and DeepSeek-R1, guided by various prompt engineering techniques, including zero-shot, few-shot, and an ensemble approach. Among all strategies, GPT-4o with prompt ensemble achieved the highest classification performance with an F1-score of 0.95 and recall of 0.98, outperforming DeepSeek-R1 on the task. The ensemble method improved reliability by aggregating outputs through embedding-based similarity and majority voting.
Chatbot Arena Meets Nuggets: Towards Explanations and Diagnostics in the Evaluation of LLM Responses
Battles, or side-by-side comparisons in so-called arenas that elicit human preferences, have emerged as a popular approach for assessing the output quality of LLMs. Recently, this idea has been extended to retrieval-augmented generation (RAG) systems. While undoubtedly representing an advance in evaluation, battles have at least two drawbacks, particularly in the context of complex information-seeking queries: they are neither explanatory nor diagnostic. Recently, the nugget evaluation methodology has emerged as a promising approach to evaluate the quality of RAG answers. Nuggets decompose long-form LLM-generated answers into atomic facts, highlighting important pieces of information necessary in a "good" response. In this work, we apply our AutoNuggetizer framework to analyze data from roughly 7K Search Arena battles provided by LMArena in a fully automatic manner. Our results show a significant correlation between nugget scores and human preferences, showcasing promise in our approach to explainable and diagnostic system evaluations. All the code necessary to reproduce results in our work is available in https://github.com/castorini/lmsys_nuggetize.
Recent advances in large language models (LLMs) demonstrate their impressive reasoning capabilities. However, the reasoning confined to internal parametric space limits LLMs' access to real-time information and understanding of the physical world. To overcome this constraint, we introduce SituatedThinker, a novel framework that enables LLMs to ground their reasoning in real-world contexts through situated thinking, which adaptively combines both internal knowledge and external information with predefined interfaces. By utilizing reinforcement learning, SituatedThinker incentivizes deliberate reasoning with the real world to acquire information and feedback, allowing LLMs to surpass their knowledge boundaries and enhance reasoning. Experimental results demonstrate significant performance improvements on multi-hop question-answering and mathematical reasoning benchmarks. Furthermore, SituatedThinker demonstrates strong performance on unseen tasks, such as KBQA, TableQA, and text-based games, showcasing the generalizable real-world grounded reasoning capability. Our codes are available at https://github.com/jnanliu/SituatedThinker.
Recent research shows that LLMs can simulate ``believable'' human behaviors to power LLM agents via prompt-only methods. In this work, we focus on evaluating LLM's objective ``accuracy'' rather than the subjective ``believability'' in simulating human behavior, leveraging a large-scale, real-world dataset collected from customers' online shopping actions. We present the first comprehensive evaluation of state-of-the-art LLMs (e.g., DeepSeek-R1, Llama, and Claude) on the task of web shopping action generation. Our results show that out-of-the-box LLM-generated actions are often misaligned with actual human behavior, whereas fine-tuning LLMs on real-world behavioral data substantially improves their ability to generate accurate actions compared to prompt-only methods. Furthermore, incorporating synthesized reasonings into model training leads to additional performance gains, demonstrating the value of explicit rationale in behavior modeling. This work evaluates state-of-the-art LLMs in behavior simulation and provides actionable insights into how real-world action data can enhance the fidelity of LLM agents.
The adoption of large language models (LLMs) as rerankers in multi-stage retrieval systems has gained significant traction in academia and industry. These models refine a candidate list of retrieved documents, often through carefully designed prompts, and are typically used in applications built on retrieval-augmented generation (RAG). This paper introduces RankLLM, an open-source Python package for reranking that is modular, highly configurable, and supports both proprietary and open-source LLMs in customized reranking workflows. To improve usability, RankLLM features optional integration with Pyserini for retrieval and provides integrated evaluation for multi-stage pipelines. Additionally, RankLLM includes a module for detailed analysis of input prompts and LLM responses, addressing reliability concerns with LLM APIs and non-deterministic behavior in Mixture-of-Experts (MoE) models. This paper presents the architecture of RankLLM, along with a detailed step-by-step guide and sample code. We reproduce results from RankGPT, LRL, RankVicuna, RankZephyr, and other recent models. RankLLM integrates with common inference frameworks and a wide range of LLMs. This compatibility allows for quick reproduction of reported results, helping to speed up both research and real-world applications. The complete repository is available at rankllm.ai, and the package can be installed via PyPI.
Understanding domain-specific theorems often requires more than just text-based reasoning; effective communication through structured visual explanations is crucial for deeper comprehension. While large language models (LLMs) demonstrate strong performance in text-based theorem reasoning, their ability to generate coherent and pedagogically meaningful visual explanations remains an open challenge. In this work, we introduce TheoremExplainAgent, an agentic approach for generating long-form theorem explanation videos (over 5 minutes) using Manim animations. To systematically evaluate multimodal theorem explanations, we propose TheoremExplainBench, a benchmark covering 240 theorems across multiple STEM disciplines, along with 5 automated evaluation metrics. Our results reveal that agentic planning is essential for generating detailed long-form videos, and the o3-mini agent achieves a success rate of 93.8% and an overall score of 0.77. However, our quantitative and qualitative studies show that most of the videos produced exhibit minor issues with visual element layout. Furthermore, multimodal explanations expose deeper reasoning flaws that text-based explanations fail to reveal, highlighting the importance of multimodal explanations.
The rapid development of LLMs has sparked extensive research into their factual knowledge. Current works find that LLMs fall short on questions around low-frequency entities. However, such proofs are unreliable since the questions can differ not only in entity frequency but also in difficulty themselves. So we introduce ComparisonQA benchmark, containing 283K abstract questions, each instantiated by a pair of high-frequency and low-frequency entities. It ensures a controllable comparison to study the role of knowledge frequency in the performance of LLMs. Because the difference between such a pair is only the entity with different frequencies. In addition, we use both correctness and uncertainty to develop a two-round method to evaluate LLMs' knowledge robustness. It aims to avoid possible semantic shortcuts which is a serious problem of current QA study. Experiments reveal that LLMs, including GPT-4o, exhibit particularly low robustness regarding low-frequency knowledge. Besides, we find that uncertainty can be used to effectively identify high-quality and shortcut-free questions while maintaining the data size. Based on this, we propose an automatic method to select such questions to form a subset called ComparisonQA-Hard, containing only hard low-frequency questions.
The emergence of large language models (LLMs) enables the development of intelligent agents capable of engaging in complex and multi-turn dialogues. However, multi-agent collaboration face critical safety challenges, such as hallucination amplification and error injection and propagation. This paper presents GUARDIAN, a unified method for detecting and mitigating multiple safety concerns in GUARDing Intelligent Agent collaboratioNs. By modeling the multi-agent collaboration process as a discrete-time temporal attributed graph, GUARDIAN explicitly captures the propagation dynamics of hallucinations and errors. The unsupervised encoder-decoder architecture incorporating an incremental training paradigm, learns to reconstruct node attributes and graph structures from latent embeddings, enabling the identification of anomalous nodes and edges with unparalleled precision. Moreover, we introduce a graph abstraction mechanism based on the Information Bottleneck Theory, which compresses temporal interaction graphs while preserving essential patterns. Extensive experiments demonstrate GUARDIAN's effectiveness in safeguarding LLM multi-agent collaborations against diverse safety vulnerabilities, achieving state-of-the-art accuracy with efficient resource utilization.
Fine-tuning Large Language Models (LLMs) for clinical Natural Language Processing (NLP) poses significant challenges due to the domain gap and limited data availability. This study investigates the effectiveness of various adapter techniques, equivalent to Low-Rank Adaptation (LoRA), for fine-tuning LLMs in a resource-constrained hospital environment. We experimented with four structures-Adapter, Lightweight, TinyAttention, and Gated Residual Network (GRN)-as final layers for clinical notes classification. We fine-tuned biomedical pre-trained models, including CamemBERT-bio, AliBERT, and DrBERT, alongside two Transformer-based models. Our extensive experimental results indicate that i) employing adapter structures does not yield significant improvements in fine-tuning biomedical pre-trained LLMs, and ii) simpler Transformer-based models, trained from scratch, perform better under resource constraints. Among the adapter structures, GRN demonstrated superior performance with accuracy, precision, recall, and an F1 score of 0.88. Moreover, the total training time for LLMs exceeded 1000 hours, compared to under 6 hours for simpler transformer-based models, highlighting that LLMs are more suitable for environments with extensive computational resources and larger datasets. Consequently, this study demonstrates that simpler Transformer-based models can be effectively trained from scratch, providing a viable solution for clinical NLP tasks in low-resource environments with limited data availability. By identifying the GRN as the most effective adapter structure, we offer a practical approach to enhance clinical note classification without requiring extensive computational resources.
Recent advances in large language models (LLMs) have enabled their use in complex agentic roles, involving decision-making with humans or other agents, making ethical alignment a key AI safety concern. While prior work has examined both LLMs' moral judgment and strategic behavior in social dilemmas, there is limited understanding of how they act when moral imperatives directly conflict with rewards or incentives. To investigate this, we introduce Moral Behavior in Social Dilemma Simulation (MoralSim) and evaluate how LLMs behave in the prisoner's dilemma and public goods game with morally charged contexts. In MoralSim, we test a range of frontier models across both game structures and three distinct moral framings, enabling a systematic examination of how LLMs navigate social dilemmas in which ethical norms conflict with payoff-maximizing strategies. Our results show substantial variation across models in both their general tendency to act morally and the consistency of their behavior across game types, the specific moral framing, and situational factors such as opponent behavior and survival risks. Crucially, no model exhibits consistently moral behavior in MoralSim, highlighting the need for caution when deploying LLMs in agentic roles where the agent's "self-interest" may conflict with ethical expectations. Our code is available at https://github.com/sbackmann/moralsim.
Large language models (LLMs) have shown promise in automating scientific hypothesis generation, yet existing approaches primarily yield coarse-grained hypotheses lacking critical methodological and experimental details. We introduce and formally define the novel task of fine-grained scientific hypothesis discovery, which entails generating detailed, experimentally actionable hypotheses from coarse initial research directions. We frame this as a combinatorial optimization problem and investigate the upper limits of LLMs' capacity to solve it when maximally leveraged. Specifically, we explore four foundational questions: (1) how to best harness an LLM's internal heuristics to formulate the fine-grained hypothesis it itself would judge as the most promising among all the possible hypotheses it might generate, based on its own internal scoring-thus defining a latent reward landscape over the hypothesis space; (2) whether such LLM-judged better hypotheses exhibit stronger alignment with ground-truth hypotheses; (3) whether shaping the reward landscape using an ensemble of diverse LLMs of similar capacity yields better outcomes than defining it with repeated instances of the strongest LLM among them; and (4) whether an ensemble of identical LLMs provides a more reliable reward landscape than a single LLM. To address these questions, we propose a hierarchical search method that incrementally proposes and integrates details into the hypothesis, progressing from general concepts to specific experimental configurations. We show that this hierarchical process smooths the reward landscape and enables more effective optimization. Empirical evaluations on a new benchmark of expert-annotated fine-grained hypotheses from recent chemistry literature show that our method consistently outperforms strong baselines.
Can LLMs accurately adjust their confidence when facing opposition? Building on previous studies measuring calibration on static fact-based question-answering tasks, we evaluate Large Language Models (LLMs) in a dynamic, adversarial debate setting, uniquely combining two realistic factors: (a) a multi-turn format requiring models to update beliefs as new information emerges, and (b) a zero-sum structure to control for task-related uncertainty, since mutual high-confidence claims imply systematic overconfidence. We organized 60 three-round policy debates among ten state-of-the-art LLMs, with models privately rating their confidence (0-100) in winning after each round. We observed five concerning patterns: (1) Systematic overconfidence: models began debates with average initial confidence of 72.9% vs. a rational 50% baseline. (2) Confidence escalation: rather than reducing confidence as debates progressed, debaters increased their win probabilities, averaging 83% by the final round. (3) Mutual overestimation: in 61.7% of debates, both sides simultaneously claimed >=75% probability of victory, a logical impossibility. (4) Persistent self-debate bias: models debating identical copies increased confidence from 64.1% to 75.2%; even when explicitly informed their chance of winning was exactly 50%, confidence still rose (from 50.0% to 57.1%). (5) Misaligned private reasoning: models' private scratchpad thoughts sometimes differed from their public confidence ratings, raising concerns about faithfulness of chain-of-thought reasoning. These results suggest LLMs lack the ability to accurately self-assess or update their beliefs in dynamic, multi-turn tasks; a major concern as LLM outputs are deployed without careful review in assistant roles or agentic settings.
We challenge the dominant focus on neural scaling laws and advocate for a paradigm shift toward downscaling in the development of large language models (LLMs). While scaling laws have provided critical insights into performance improvements through increasing model and dataset size, we emphasize the significant limitations of this approach, particularly in terms of computational inefficiency, environmental impact, and deployment constraints. To address these challenges, we propose a holistic framework for downscaling LLMs that seeks to maintain performance while drastically reducing resource demands. This paper outlines practical strategies for transitioning away from traditional scaling paradigms, advocating for a more sustainable, efficient, and accessible approach to LLM development.
LLM-as-a-Judge employs large language models (LLMs), such as GPT-4, to evaluate the quality of LLM-generated responses, gaining popularity for its cost-effectiveness and strong alignment with human evaluations. However, training proxy judge models using evaluation data generated by powerful teacher models introduces a critical yet previously overlooked issue: teacher preference bias, where the proxy judge model learns a biased preference for responses from the teacher model. To tackle this problem, we propose a novel setting that incorporates an additional assistant model, which is not biased toward the teacher model's responses, to complement the training data. Building on this setup, we introduce AGDe-Judge, a three-stage framework designed to debias from both the labels and feedbacks in the training data. Extensive experiments demonstrate that AGDe-Judge effectively reduces teacher preference bias while maintaining strong performance across six evaluation benchmarks. Code is available at https://github.com/Liuz233/AGDe-Judge.
Effective teaching requires adapting instructional strategies to accommodate the diverse cognitive and behavioral profiles of students, a persistent challenge in education and teacher training. While Large Language Models (LLMs) offer promise as tools to simulate such complex pedagogical environments, current simulation frameworks are limited in two key respects: (1) they often reduce students to static knowledge profiles, and (2) they lack adaptive mechanisms for modeling teachers who evolve their strategies in response to student feedback. To address these gaps, \textbf{we introduce a novel simulation framework that integrates LLM-based heterogeneous student agents with a self-optimizing teacher agent}. The teacher agent's pedagogical policy is dynamically evolved using a genetic algorithm, allowing it to discover and refine effective teaching strategies based on the aggregate performance of diverse learners. In addition, \textbf{we propose Persona-RAG}, a Retrieval Augmented Generation module that enables student agents to retrieve knowledge tailored to their individual learning styles. Persona-RAG preserves the retrieval accuracy of standard RAG baselines while enhancing personalization, an essential factor in modeling realistic educational scenarios. Through extensive experiments, we demonstrate how our framework supports the emergence of distinct and interpretable teaching patterns when interacting with varied student populations. Our results highlight the potential of LLM-driven simulations to inform adaptive teaching practices and provide a testbed for training human educators in controlled, data-driven environments.
Role-based access control (RBAC) and hierarchical structures are foundational to how information flows and decisions are made within virtually all organizations. As the potential of Large Language Models (LLMs) to serve as unified knowledge repositories and intelligent assistants in enterprise settings becomes increasingly apparent, a critical, yet under explored, challenge emerges: \textit{can these models reliably understand and operate within the complex, often nuanced, constraints imposed by organizational hierarchies and associated permissions?} Evaluating this crucial capability is inherently difficult due to the proprietary and sensitive nature of real-world corporate data and access control policies. We introduce a synthetic yet representative \textbf{OrgAccess} benchmark consisting of 40 distinct types of permissions commonly relevant across different organizational roles and levels. We further create three types of permissions: 40,000 easy (1 permission), 10,000 medium (3-permissions tuple), and 20,000 hard (5-permissions tuple) to test LLMs' ability to accurately assess these permissions and generate responses that strictly adhere to the specified hierarchical rules, particularly in scenarios involving users with overlapping or conflicting permissions. Our findings reveal that even state-of-the-art LLMs struggle significantly to maintain compliance with role-based structures, even with explicit instructions, with their performance degrades further when navigating interactions involving two or more conflicting permissions. Specifically, even \textbf{GPT-4.1 only achieves an F1-Score of 0.27 on our hardest benchmark}. This demonstrates a critical limitation in LLMs' complex rule following and compositional reasoning capabilities beyond standard factual or STEM-based benchmarks, opening up a new paradigm for evaluating their fitness for practical, structured environments.
Large Language Models (LLMs) have demonstrated remarkable performance across various disciplines and tasks. However, benchmarking their capabilities with multilingual spoken queries remains largely unexplored. In this study, we introduce SpokenNativQA, the first multilingual and culturally aligned spoken question-answering (SQA) dataset designed to evaluate LLMs in real-world conversational settings. The dataset comprises approximately 33,000 naturally spoken questions and answers in multiple languages, including low-resource and dialect-rich languages, providing a robust benchmark for assessing LLM performance in speech-based interactions. SpokenNativQA addresses the limitations of text-based QA datasets by incorporating speech variability, accents, and linguistic diversity. We benchmark different ASR systems and LLMs for SQA and present our findings. We released the data at (https://huggingface.co/datasets/QCRI/SpokenNativQA) and the experimental scripts at (https://llmebench.qcri.org/) for the research community.
Due to the auto-regressive nature of current video large language models (Video-LLMs), the inference latency increases as the input sequence length grows, posing challenges for the efficient processing of video sequences that are usually very long. We observe that during decoding, the attention scores of most tokens in Video-LLMs tend to be sparse and concentrated, with only certain tokens requiring comprehensive full attention. Based on this insight, we introduce Sparse-to-Dense (StD), a novel decoding strategy that integrates two distinct modules: one leveraging sparse top-K attention and the other employing dense full attention. These modules collaborate to accelerate Video-LLMs without loss. The fast (sparse) model speculatively decodes multiple tokens, while the slow (dense) model verifies them in parallel. StD is a tuning-free, plug-and-play solution that achieves up to a 1.94$\times$ walltime speedup in video processing. It maintains model performance while enabling a seamless transition from a standard Video-LLM to a sparse Video-LLM with minimal code modifications.
Large Language Models (LLMs) often struggle with data-analytics requests related to information retrieval and data manipulation that frequently arise in real-world scenarios under multiple conditions. In this paper, we introduce Thinking with Tables, where we inject tabular structures into LLMs for data-analytics requests. Through comprehensive evaluations across various request types, we show that providing tabular structures yields a 40.29 percent average performance gain along with better robustness and token efficiency. Through attention-value analysis, we uncover that tables help LLMs better attend to relevant information, explaining these improvements. Beyond tables and text, we evaluate whether (1) blending structuredness within text, such as providing templates or fixing the order of attributes, and (2) other representative structures, such as knowledge graphs and JSON, are helpful. We observe that utilizing tables offers the best balance between efficiency and effectiveness. These advantages remain consistent under increased task complexity and even when all input data cannot be structured. Finally, as data analytics typically relies on structured factual inputs, our text-to-table conversion demonstrates the method's applicability to text-compatible data sources.
Ensuring the reliability of Large Language Models (LLMs) in complex reasoning tasks remains a formidable challenge, particularly in scenarios that demand precise mathematical calculations and knowledge-intensive open-domain generation. In this work, we introduce an uncertainty-aware framework designed to enhance the accuracy of LLM responses by systematically incorporating model confidence at critical decision points. We propose an approach that encourages multi-step reasoning in LLMs and quantify the confidence of intermediate answers such as numerical results in mathematical reasoning and proper nouns in open-domain generation. Then, the overall confidence of each reasoning chain is evaluated based on confidence of these critical intermediate steps. Finally, we aggregate the answer of generated response paths in a way that reflects the reliability of each generated content (as opposed to self-consistency in which each generated chain contributes equally to majority voting). We conducted extensive experiments in five datasets, three mathematical datasets and two open-domain datasets, using four LLMs. The results consistently validate the effectiveness of our novel confidence aggregation method, leading to an accuracy improvement of up to 7.4% and 5.8% over baseline approaches in math and open-domain generation tasks, respectively. Code is publicly available at https://github.com/ Aquasar11/CER.
We demonstrate, for the first time, fully quantized training (FQT) of large language models (LLMs) using predominantly 4-bit floating-point (FP4) precision for weights, activations, and gradients on datasets up to 200 billion tokens. We extensively investigate key design choices for FP4, including block sizes, scaling formats, and rounding methods. Our analysis shows that the NVFP4 format, where each block of 16 FP4 values (E2M1) shares a scale represented in E4M3, provides optimal results. We use stochastic rounding for backward and update passes and round-to-nearest for the forward pass to enhance stability. Additionally, we identify a theoretical and empirical threshold for effective quantized training: when the gradient norm falls below approximately $\sqrt{3}$ times the quantization noise, quantized training becomes less effective. Leveraging these insights, we successfully train a 7-billion-parameter model on 256 Intel Gaudi2 accelerators. The resulting FP4-trained model achieves downstream task performance comparable to a standard BF16 baseline, confirming that FP4 training is a practical and highly efficient approach for large-scale LLM training. A reference implementation is supplied in https://github.com/Anonymous1252022/fp4-all-the-way .
Large Language Models (LLMs) have demonstrated impressive reasoning capabilities in complex problem-solving tasks, sparking growing interest in their application to preference reasoning in recommendation systems. Existing methods typically rely on fine-tuning with explicit chain-of-thought (CoT) data. However, these methods face significant practical limitations due to (1) the difficulty of obtaining high-quality CoT data in recommendation and (2) the high inference latency caused by generating CoT reasoning. In this work, we explore an alternative approach that shifts from explicit CoT reasoning to compact, information-dense latent reasoning. This approach eliminates the need for explicit CoT generation and improves inference efficiency, as a small set of latent tokens can effectively capture the entire reasoning process. Building on this idea, we propose $\textit{\underline{R}einforced \underline{Latent} \underline{R}easoning for \underline{R}ecommendation}$ (LatentR$^3$), a novel end-to-end training framework that leverages reinforcement learning (RL) to optimize latent reasoning without relying on any CoT data.LatentR$^3$ adopts a two-stage training strategy: first, supervised fine-tuning to initialize the latent reasoning module, followed by pure RL training to encourage exploration through a rule-based reward design. Our RL implementation is based on a modified GRPO algorithm, which reduces computational overhead during training and introduces continuous reward signals for more efficient learning. Extensive experiments demonstrate that LatentR$^3$ enables effective latent reasoning without any direct supervision of the reasoning process, significantly improving performance when integrated with different LLM-based recommendation methods. Our codes are available at https://anonymous.4open.science/r/R3-A278/.
The rapid advancement of scientific progress requires innovative tools that can accelerate knowledge discovery. Although recent AI methods, particularly large language models (LLMs), have shown promise in tasks such as hypothesis generation and experimental design, they fall short of replicating the collaborative nature of real-world scientific practices, where diverse experts work together in teams to tackle complex problems. To address the limitations, we propose an LLM-based multi-agent system, i.e., Virtual Scientists (VirSci), designed to mimic the teamwork inherent in scientific research. VirSci organizes a team of agents to collaboratively generate, evaluate, and refine research ideas. Through comprehensive experiments, we demonstrate that this multi-agent approach outperforms the state-of-the-art method in producing novel scientific ideas. We further investigate the collaboration mechanisms that contribute to its tendency to produce ideas with higher novelty, offering valuable insights to guide future research and illuminating pathways toward building a robust system for autonomous scientific discovery. The code is available at https://github.com/open-sciencelab/Virtual-Scientists.
Large Language Models (LLMs) have demonstrated remarkable general capabilities, but enhancing skills such as reasoning often demands substantial computational resources and may compromise their generalization. While Parameter-Efficient Fine-Tuning (PEFT) methods offer a more resource-conscious alternative, they typically requires retraining for each LLM backbone due to architectural dependencies. To address these challenges, here we propose Universal Reasoner (UniR) - a single, lightweight, composable, and plug-and-play reasoning module that can be used with any frozen LLM to endow it with specialized reasoning capabilities. Specifically, UniR decomposes the reward into a standalone reasoning module that is trained independently using predefined rewards, effectively translating trajectory-level signals into token-level guidance. Once trained, UniR can be combined with any frozen LLM at inference time by simply adding its output logits to those of the LLM backbone. This additive structure naturally enables modular composition: multiple UniR modules trained for different tasks can be jointly applied by summing their logits, enabling complex reasoning via composition. Experimental results on mathematical reasoning and machine translation tasks show that UniR significantly outperforms \add{existing baseline fine-tuning methods using the Llama3.2 model}. Furthermore, UniR demonstrates strong weak-to-strong generalization: reasoning modules trained on smaller models effectively guide much larger LLMs. This makes UniR a cost-efficient, adaptable, and robust solution for enhancing reasoning in LLMs without compromising their core capabilities. Code is open-sourced at https://github.com/hangeol/UniR
Intelligent tutoring agents powered by large language models (LLMs) have been increasingly explored to deliver personalized knowledge in areas such as language learning and science education. However, their capabilities in guiding users to solve complex real-world tasks remain underexplored. To address this limitation, in this work, we focus on coding tutoring, a challenging problem that requires tutors to proactively guide students towards completing predefined coding tasks. We propose a novel agent workflow, Trace-and-Verify (TRAVER), which combines knowledge tracing to estimate a student's knowledge state and turn-by-turn verification to ensure effective guidance toward task completion. We introduce DICT, an automatic evaluation protocol that assesses tutor agents using controlled student simulation and code generation tests. Extensive experiments reveal the challenges of coding tutoring and demonstrate that TRAVER achieves a significantly higher success rate. Although we use code tutoring as an example in this paper, our approach can be extended beyond coding, providing valuable insights into advancing tutoring agents for human task learning.
Large language models (LLMs) are typically aligned to comply with safety guidelines by refusing harmful instructions. A recent attack, termed abliteration, isolates and suppresses the single latent direction most responsible for refusal behavior, enabling the model to generate unethical content. We propose a defense that modifies how models generate refusals. We construct an extended-refusal dataset that contains harmful prompts with a full response that justifies the reason for refusal. We then fine-tune Llama-2-7B-Chat and Qwen2.5-Instruct (1.5B and 3B parameters) on our extended-refusal dataset, and evaluate the resulting systems on a set of harmful prompts. In our experiments, extended-refusal models maintain high refusal rates, dropping at most by 10%, whereas baseline models' refusal rates drop by 70-80% after abliteration. A broad evaluation of safety and utility shows that extended-refusal fine-tuning neutralizes the abliteration attack while preserving general performance.
In this work, we introduce PII-Scope, a comprehensive benchmark designed to evaluate state-of-the-art methodologies for PII extraction attacks targeting LLMs across diverse threat settings. Our study provides a deeper understanding of these attacks by uncovering several hyperparameters (e.g., demonstration selection) crucial to their effectiveness. Building on this understanding, we extend our study to more realistic attack scenarios, exploring PII attacks that employ advanced adversarial strategies, including repeated and diverse querying, and leveraging iterative learning for continual PII extraction. Through extensive experimentation, our results reveal a notable underestimation of PII leakage in existing single-query attacks. In fact, we show that with sophisticated adversarial capabilities and a limited query budget, PII extraction rates can increase by up to fivefold when targeting the pretrained model. Moreover, we evaluate PII leakage on finetuned models, showing that they are more vulnerable to leakage than pretrained models. Overall, our work establishes a rigorous empirical benchmark for PII extraction attacks in realistic threat scenarios and provides a strong foundation for developing effective mitigation strategies.
Large language models (LLMs) are increasingly expected to tackle complex tasks, driven by their expanding applications and users' growing proficiency in crafting sophisticated prompts. However, as the number of explicitly stated requirements increases (particularly more than 10 constraints), LLMs often struggle to accurately follow such complex instructions. To address this challenge, we propose RECAST, a novel framework for synthesizing datasets where each example incorporates far more constraints than those in existing benchmarks. These constraints are extracted from real-world prompt-response pairs to ensure practical relevance. RECAST enables automatic verification of constraint satisfaction via rule-based validators for quantitative constraints and LLM-based validators for qualitative ones. Using this framework, we construct RECAST-30K, a large-scale, high-quality dataset comprising 30k instances spanning 15 constraint types. Experimental results demonstrate that models fine-tuned on RECAST-30K show substantial improvements in following complex instructions. Moreover, the verifiability provided by RECAST enables the design of reward functions for reinforcement learning, which further boosts model performance on complex and challenging tasks.
Accurate uncertainty quantification in large language models (LLMs) is essential for providing credible confidence estimates over their outputs. However, fine-tuned LLMs often exhibit overconfidence in uncertain predictions, which stems from their limited ability to generalize with sparse data. Existing parameter efficient fine-tuning (PEFT) uncertainty quantification methods for LLMs focus on post fine-tuning stage, and thus fail to address the core issue: limited specialization of PEFT adapters to accurately capture task-specific input-output relationships. To address these limitations, we propose Functional-Level Uncertainty Quantification for Calibrated Fine-Tuning (UQ4CT), which captures and calibrates uncertainty over the space of functions that map input prompts to outputs. We implement UQ4CT during the fine-tuning stage via a mixture-of-experts framework that hierarchically decomposes the functional space. Empirically, UQ4CT achieves over $25\%$ reduction in Expected Calibration Error (ECE) while preserving high accuracy across five benchmarks. Even under distribution shift, UQ4CT maintains superior ECE performance with high accuracy, showcasing improved generalizability.
The advent of large language models (LLMs) presents new opportunities for travel demand modeling. However, behavioral misalignment between LLMs and humans presents obstacles for the usage of LLMs, and existing alignment methods are frequently inefficient or impractical given the constraints of typical travel demand data. This paper introduces a novel framework for aligning LLMs with human travel choice behavior, tailored to the current travel demand data sources. Our framework uses a persona inference and loading process to condition LLMs with suitable prompts to enhance alignment. The inference step establishes a set of base personas from empirical data, and a learned persona loading function driven by behavioral embeddings guides the loading process. We validate our framework on the Swissmetro mode choice dataset, and the results show that our proposed approach significantly outperformed baseline choice models and LLM-based simulation models in predicting both aggregate mode choice shares and individual choice outcomes. Furthermore, we showcase that our framework can generate insights on population behavior through interpretable parameters. Overall, our research offers a more adaptable, interpretable, and resource-efficient pathway to robust LLM-based travel behavior simulation, paving the way to integrate LLMs into travel demand modeling practice in the future.
Applying Reinforcement Learning (RL) to Video Large Language Models (Video-LLMs) shows significant promise for complex video reasoning. However, popular Reinforcement Fine-Tuning (RFT) methods, such as outcome-based Group Relative Policy Optimization (GRPO), are limited by data preparation bottlenecks (e.g., noise or high cost) and exhibit unstable improvements in the quality of long chain-of-thoughts (CoTs) and downstream performance.To address these limitations, we propose VerIPO, a Verifier-guided Iterative Policy Optimization method designed to gradually improve video LLMs' capacity for generating deep, long-term reasoning chains. The core component is Rollout-Aware Verifier, positioned between the GRPO and Direct Preference Optimization (DPO) training phases to form the GRPO-Verifier-DPO training loop. This verifier leverages small LLMs as a judge to assess the reasoning logic of rollouts, enabling the construction of high-quality contrastive data, including reflective and contextually consistent CoTs. These curated preference samples drive the efficient DPO stage (7x faster than GRPO), leading to marked improvements in reasoning chain quality, especially in terms of length and contextual consistency. This training loop benefits from GRPO's expansive search and DPO's targeted optimization. Experimental results demonstrate: 1) Significantly faster and more effective optimization compared to standard GRPO variants, yielding superior performance; 2) Our trained models exceed the direct inference of large-scale instruction-tuned Video-LLMs, producing long and contextually consistent CoTs on diverse video reasoning tasks; and 3) Our model with one iteration outperforms powerful LMMs (e.g., Kimi-VL) and long reasoning models (e.g., Video-R1), highlighting its effectiveness and stability.
Safety concerns in large language models (LLMs) have gained significant attention due to their exposure to potentially harmful data during pre-training. In this paper, we identify a new safety vulnerability in LLMs: their susceptibility to \textit{natural distribution shifts} between attack prompts and original toxic prompts, where seemingly benign prompts, semantically related to harmful content, can bypass safety mechanisms. To explore this issue, we introduce a novel attack method, \textit{ActorBreaker}, which identifies actors related to toxic prompts within pre-training distribution to craft multi-turn prompts that gradually lead LLMs to reveal unsafe content. ActorBreaker is grounded in Latour's actor-network theory, encompassing both human and non-human actors to capture a broader range of vulnerabilities. Our experimental results demonstrate that ActorBreaker outperforms existing attack methods in terms of diversity, effectiveness, and efficiency across aligned LLMs. To address this vulnerability, we propose expanding safety training to cover a broader semantic space of toxic content. We thus construct a multi-turn safety dataset using ActorBreaker. Fine-tuning models on our dataset shows significant improvements in robustness, though with some trade-offs in utility. Code is available at https://github.com/AI45Lab/ActorAttack.
Despite the success of Large Language Models (LLMs) across various fields, their potential to generate untruthful, biased and harmful responses poses significant risks, particularly in critical applications. This highlights the urgent need for systematic methods to detect and prevent such misbehavior. While existing approaches target specific issues such as harmful responses, this work introduces LLMScan, an innovative LLM monitoring technique based on causality analysis, offering a comprehensive solution. LLMScan systematically monitors the inner workings of an LLM through the lens of causal inference, operating on the premise that the LLM's `brain' behaves differently when misbehaving. By analyzing the causal contributions of the LLM's input tokens and transformer layers, LLMScan effectively detects misbehavior. Extensive experiments across various tasks and models reveal clear distinctions in the causal distributions between normal behavior and misbehavior, enabling the development of accurate, lightweight detectors for a variety of misbehavior detection tasks.
Bilevel optimization has shown its utility across various machine learning settings, yet most algorithms in practice require second-order information, making it challenging to scale them up. Only recently, a paradigm of first-order algorithms has emerged in the theoretical literature, capable of effectively addressing bilevel optimization problems. Nevertheless, the practical efficiency of this paradigm remains unverified, particularly in the context of large language models (LLMs). This paper introduces the first scalable instantiation of this paradigm called ScaleBiO, focusing on bilevel optimization for large-scale LLM data reweighting. By combining with a recently proposed memory-efficient training technique called LISA, our novel algorithm allows the paradigm to scale to $\sim$30B-sized LLMs on $8\times$H100 GPUs, marking the first successful application of bilevel optimization under practical scenarios for large-sized LLMs. Empirically, extensive experiments on data reweighting verify the effectiveness of ScaleBiO for different-scaled models, including Llama-3-8B, Gemma-2-9B, Qwen-2-7B, and Qwen-2.5-32B, where bilevel optimization succeeds in instruction-following and math reasoning tasks, outperforming several popular baselines, including uniform sampling, influence-aware data filtering, and reference-model-based sampling methods. Theoretically, ScaleBiO ensures the optimality of the learned data weights, along with a convergence guarantee matching the conventional first-order bilevel optimization paradigm on smooth and strongly convex objectives.
Large language models (LLMs) have demonstrated impressive performance on natural language tasks, but their decision-making processes remain largely opaque. Existing explanation methods either suffer from limited faithfulness to the model's reasoning or produce explanations that humans find difficult to understand. To address these challenges, we propose \textbf{ProtoSurE}, a novel prototype-based surrogate framework that provides faithful and human-understandable explanations for LLMs. ProtoSurE trains an interpretable-by-design surrogate model that aligns with the target LLM while utilizing sentence-level prototypes as human-understandable concepts. Extensive experiments show that ProtoSurE consistently outperforms SOTA explanation methods across diverse LLMs and datasets. Importantly, ProtoSurE demonstrates strong data efficiency, requiring relatively few training examples to achieve good performance, making it practical for real-world applications.
Compressing long chain-of-thought (CoT) from large language models (LLMs) is an emerging strategy to improve the reasoning efficiency of LLMs. Despite its promising benefits, existing studies equally compress all thoughts within a long CoT, hindering more concise and effective reasoning. To this end, we first investigate the importance of different thoughts by examining their effectiveness and efficiency in contributing to reasoning through automatic long CoT chunking and Monte Carlo rollouts. Building upon the insights, we propose a theoretically bounded metric to jointly measure the effectiveness and efficiency of different thoughts. We then propose Long$\otimes$Short, an efficient reasoning framework that enables two LLMs to collaboratively solve the problem: a long-thought LLM for more effectively generating important thoughts, while a short-thought LLM for efficiently generating remaining thoughts. Specifically, we begin by synthesizing a small amount of cold-start data to fine-tune LLMs for long-thought and short-thought reasoning styles, respectively. Furthermore, we propose a synergizing-oriented multi-turn reinforcement learning, focusing on the model self-evolution and collaboration between long-thought and short-thought LLMs. Experimental results show that our method enables Qwen2.5-7B and Llama3.1-8B to achieve comparable performance compared to DeepSeek-R1-Distill-Qwen-7B and DeepSeek-R1-Distill-Llama-8B, while reducing token length by over 80% across the MATH500, AIME24/25, AMC23, and GPQA Diamond benchmarks. Our data and code are available at https://github.com/usail-hkust/LongShort.
Penetration testing is a critical technique for identifying security vulnerabilities, traditionally performed manually by skilled security specialists. This complex process involves gathering information about the target system, identifying entry points, exploiting the system, and reporting findings. Despite its effectiveness, manual penetration testing is time-consuming and expensive, often requiring significant expertise and resources that many organizations cannot afford. While automated penetration testing methods have been proposed, they often fall short in real-world applications due to limitations in flexibility, adaptability, and implementation. Recent advancements in large language models (LLMs) offer new opportunities for enhancing penetration testing through increased intelligence and automation. However, current LLM-based approaches still face significant challenges, including limited penetration testing knowledge and a lack of comprehensive automation capabilities. To address these gaps, we propose PentestAgent, a novel LLM-based automated penetration testing framework that leverages the power of LLMs and various LLM-based techniques like Retrieval Augmented Generation (RAG) to enhance penetration testing knowledge and automate various tasks. Our framework leverages multi-agent collaboration to automate intelligence gathering, vulnerability analysis, and exploitation stages, reducing manual intervention. We evaluate PentestAgent using a comprehensive benchmark, demonstrating superior performance in task completion and overall efficiency. This work significantly advances the practical applicability of automated penetration testing systems.
Querying tables with unstructured data is challenging due to the presence of text (or image), either embedded in the table or in external paragraphs, which traditional SQL struggles to process, especially for tasks requiring semantic reasoning. While Large Language Models (LLMs) excel at understanding context, they face limitations with long input sequences. Existing approaches that combine SQL and LLMs typically rely on rigid, predefined work-flows, limiting their adaptability to complex queries. To address these issues, we introduce Weaver , a modular pipeline that dynamically integrates SQL and LLMs for table-based question answering (TableQA). Weaver generates a flexible, step-by-step plan that combines SQL for structured data retrieval with LLMs for semantic processing. By decomposing complex queries into manageable subtasks, Weaver improves accuracy and generalization. Our experiments show that Weaver consistently outperforms state-of-the-art methods across four TableQA datasets, reducing both API calls and error rates.
Instruction-tuned large language models (LLMs) employ structured templates, such as role markers and special tokens, to enforce format consistency during inference. However, we identify a critical limitation of such formatting: it induces a phenomenon we term diversity collapse, where the model generates semantically similar outputs for open-ended inputs, undermining creativity and variability. We systematically evaluate this effect across tasks like story completion and free-form generation, finding that (1) diversity collapse persists even under high-temperature sampling, and (2) structural tokens in templates significantly constrain the model's output space. To contextualize these findings, we fine-tune the same model using a range of structured prompts and then evaluate them across three axes: downstream task performance, alignment behavior, and output diversity. Our analysis shows that format consistency between fine-tuning and inference is crucial for structure-sensitive tasks (e.g., GSM8K, IFEval), but has marginal influence on knowledge-heavy tasks (e.g., MMLU, WebQuestions). In contrast, output diversity is primarily governed by the presence or absence of structural tokens, with minimal formatting yielding the most diverse outputs. These findings reveal that current prompting conventions, while beneficial for alignment, may inadvertently suppress output diversity, underscoring the need for diversity-aware prompt design and instruction tuning.
Despite LLMs' explicit alignment against demographic stereotypes, they have been shown to exhibit biases under various social contexts. In this work, we find that LLMs exhibit concerning biases in how they associate solution veracity with demographics. Through experiments across five human value-aligned LLMs on mathematics, coding, commonsense, and writing problems, we reveal two forms of such veracity biases: Attribution Bias, where models disproportionately attribute correct solutions to certain demographic groups, and Evaluation Bias, where models' assessment of identical solutions varies based on perceived demographic authorship. Our results show pervasive biases: LLMs consistently attribute fewer correct solutions and more incorrect ones to African-American groups in math and coding, while Asian authorships are least preferred in writing evaluation. In additional studies, we show LLMs automatically assign racially stereotypical colors to demographic groups in visualization code, suggesting these biases are deeply embedded in models' reasoning processes. Our findings indicate that demographic bias extends beyond surface-level stereotypes and social context provocations, raising concerns about LLMs' deployment in educational and evaluation settings.
Large language models (LLMs) continue to face challenges in reliably solving reasoning tasks, particularly those that require precise rule following, as often found in mathematical reasoning. This paper introduces a novel neurosymbolic method that improves LLM reasoning by encoding hidden states into neurosymbolic vectors, enabling problem-solving within a neurosymbolic vector space. The results are decoded and merged with the original hidden state, significantly boosting the model's performance on numerical reasoning tasks. By offloading computation through neurosymbolic representations, this method enhances efficiency, reliability, and interpretability. Experimental results demonstrate an average of 88.6% lower cross-entropy loss and 15.4 times more problems correctly solved on a suite of mathematical reasoning tasks compared to chain-of-thought prompting and supervised fine-tuning (LoRA), without degrading performance on other tasks. We make our code available at: https://github.com/vdhanraj/Neurosymbolic-LLM.
Traditional recommender systems usually take the user-platform paradigm, where users are directly exposed under the control of the platform's recommendation algorithms. However, the defect of recommendation algorithms may put users in very vulnerable positions under this paradigm. First, many sophisticated models are often designed with commercial objectives in mind, focusing on the platform's benefits, which may hinder their ability to protect and capture users' true interests. Second, these models are typically optimized using data from all users, which may overlook individual user's preferences. Due to these shortcomings, users may experience several disadvantages under the traditional user-platform direct exposure paradigm, such as lack of control over the recommender system, potential manipulation by the platform, echo chamber effects, or lack of personalization for less active users due to the dominance of active users during collaborative learning. Therefore, there is an urgent need to develop a new paradigm to protect user interests and alleviate these issues. Recently, some researchers have introduced LLM agents to simulate user behaviors, these approaches primarily aim to optimize platform-side performance, leaving core issues in recommender systems unresolved. To address these limitations, we propose a new user-agent-platform paradigm, where agent serves as the protective shield between user and recommender system that enables indirect exposure.
We present a novel active learning framework for 3D point cloud semantic segmentation that, for the first time, integrates large language models (LLMs) to construct hierarchical label structures and guide uncertainty-based sample selection. Unlike prior methods that treat labels as flat and independent, our approach leverages LLM prompting to automatically generate multi-level semantic taxonomies and introduces a recursive uncertainty projection mechanism that propagates uncertainty across hierarchy levels. This enables spatially diverse, label-aware point selection that respects the inherent semantic structure of 3D scenes. Experiments on S3DIS and ScanNet v2 show that our method achieves up to 4% mIoU improvement under extremely low annotation budgets (e.g., 0.02%), substantially outperforming existing baselines. Our results highlight the untapped potential of LLMs as knowledge priors in 3D vision and establish hierarchical uncertainty modeling as a powerful paradigm for efficient point cloud annotation.
To alleviate the memory bandwidth bottleneck in Large Language Model (LLM) inference workloads, weight matrices are stored in memory in quantized and sparsified formats. Hence, before tiles of these matrices can be processed by in-core generalized matrix multiplication (GeMM) hardware engines, they need to be dequantized and de-sparsified. This is currently performed in software with vector operations. Unfortunately, this approach delivers only modest performance. Moreover, it is hard to understand how to improve the system, as the overall GeMM performance depends on the interaction between memory resources, vector units, and hardware matrix engines. To improve the performance of LLM inference in advanced platforms equipped with in-core GeMM engines and HBM, this paper makes three main contributions. First, it develops an analytical performance model with a 3D visual representation that provides insights into how memory resources, vector units, and hardware matrix engines interact to deliver compressed GeMM performance. Second, it proposes DECA, a new near-core ML-model decompression accelerator. DECA offloads tile de-sparsification and dequantization from the CPU, producing ready-to-use tiles for in-core GeMM engines. Third, it introduces a new ISA extension that enables out-of-order invocation of the near-core accelerator. With this extension, accelerator and core computations can interleave and overlap with high-performance. Our evaluation shows that, in a simulated 56-core Xeon 4 server with HBM, DECA accelerates the execution of compressed GeMMs by up to 4x over the use of optimized Intel software kernels. Further, DECA reduces the next-token generation time of Llama2-70B and OPT-66B by 1.6x-2.6x.
Natural language generation (NLG) metrics play a central role in evaluating generated texts, but are not well suited for the structural and legal characteristics of patent documents. Large language models (LLMs) offer strong potential in automating patent generation, yet research on evaluating LLM-generated patents remains limited, especially in evaluating the generation quality of patent claims, which are central to defining the scope of protection. Effective claim evaluation requires addressing legal validity, technical accuracy, and structural compliance. To address this gap, we introduce PatentScore, a multi-dimensional evaluation framework for assessing LLM-generated patent claims. PatentScore incorporates: (1) hierarchical decomposition for claim analysis; (2) domain-specific validation patterns based on legal and technical standards; and (3) scoring across structural, semantic, and legal dimensions. Unlike general-purpose NLG metrics, PatentScore reflects patent-specific constraints and document structures, enabling evaluation beyond surface similarity. We evaluate 400 GPT-4o-mini generated Claim 1s and report a Pearson correlation of $r = 0.819$ with expert annotations, outperforming existing NLG metrics. Furthermore, we conduct additional evaluations using open models such as Claude-3.5-Haiku and Gemini-1.5-flash, all of which show strong correlations with expert judgments, confirming the robustness and generalizability of our framework.
Large language models (LLMs) obtain state of the art zero shot relevance ranking performance on a variety of information retrieval tasks. The two most common prompts to elicit LLM relevance judgments are pointwise scoring (a.k.a. relevance generation), where the LLM sees a single query-document pair and outputs a single relevance score, and listwise ranking (a.k.a. permutation generation), where the LLM sees a query and a list of documents and outputs a permutation, sorting the documents in decreasing order of relevance. The current research community consensus is that listwise ranking yields superior performance, and significant research effort has been devoted to crafting LLM listwise ranking algorithms. The underlying hypothesis is that LLMs are better at making relative relevance judgments than absolute ones. In tension with this hypothesis, we find that the gap between pointwise scoring and listwise ranking shrinks when pointwise scoring is implemented using a sufficiently large ordinal relevance label space, becoming statistically insignificant for many LLM-benchmark dataset combinations (where ``significant'' means ``95\% confidence that listwise ranking improves NDCG@10''). Our evaluations span four LLMs, eight benchmark datasets from the BEIR and TREC-DL suites, and two proprietary datasets with relevance labels collected after the training cut-off of all LLMs evaluated.
Foundation models based on large language models (LLMs) have shown great success in handling various tasks and modalities. However, adapting these models for general-purpose audio-language tasks is challenging due to differences in acoustic environments and task variations. In this work, we introduce LiSTEN Learning Soft Token Embeddings for Neural Audio LLMs), a framework for adapting LLMs to speech and audio tasks. LiSTEN uses a dynamic prompt selection strategy with learnable key-value pairs, allowing the model to balance general and task-specific knowledge while avoiding overfitting in a multitask setting. Our approach reduces dependence on large-scale ASR or captioning datasets, achieves competitive performance with fewer trainable parameters, and simplifies training by using a single-stage process. Additionally, LiSTEN enhances interpretability by analyzing the diversity and overlap of selected prompts across different tasks.
Heterogeneous device-edge-cloud computing infrastructures have become widely adopted in telecommunication operators and Wide Area Networks (WANs), offering multi-tier computational support for emerging intelligent services. With the rapid proliferation of Large Language Model (LLM) services, efficiently coordinating inference tasks and reducing communication overhead within these multi-tier network architectures becomes a critical deployment challenge. Existing LLM serving paradigms exhibit significant limitations: on-device deployment supports only lightweight LLMs due to hardware constraints, while cloud-centric deployment suffers from resource congestion and considerable prompt communication overhead caused by frequent service requests during peak periods. Although the model-cascading-based inference strategy adapts better to multi-tier networks, its reliance on fine-grained, manually adjusted thresholds makes it less responsive to dynamic network conditions and varying task complexities. To address these challenges, we propose RecServe, a recursive offloading framework tailored for LLM serving in multi-tier networks. RecServe integrates a task-specific hierarchical confidence evaluation mechanism that guides offloading decisions based on inferred task complexity in progressively scaled LLMs across device, edge, and cloud tiers. To further enable intelligent task routing across tiers, RecServe employs a sliding-window-based dynamic offloading strategy with quantile interpolation, enabling real-time tracking of historical confidence distributions and adaptive offloading threshold adjustments. Experiments on eight datasets demonstrate that RecServe outperforms CasServe in both service quality and communication efficiency, and reduces the communication burden by over 50\% compared to centralized cloud-based serving.
Although Large Language Models (LLMs) have demonstrated remarkable progress, their proficiency in graph-related tasks remains notably limited, hindering the development of truly general-purpose models. Previous attempts, including pretraining graph foundation models or employing supervised fine-tuning, often face challenges such as the scarcity of large-scale, universally represented graph data. We introduce G1, a simple yet effective approach demonstrating that Reinforcement Learning (RL) on synthetic graph-theoretic tasks can significantly scale LLMs' graph reasoning abilities. To enable RL training, we curate Erd\~os, the largest graph reasoning dataset to date comprising 50 diverse graph-theoretic tasks of varying difficulty levels, 100k training data and 5k test data, all drived from real-world graphs. With RL on Erd\~os, G1 obtains substantial improvements in graph reasoning, where our finetuned 3B model even outperforms Qwen2.5-72B-Instruct (24x size). RL-trained models also show strong zero-shot generalization to unseen tasks, domains, and graph encoding schemes, including other graph-theoretic benchmarks as well as real-world node classification and link prediction tasks, without compromising general reasoning abilities. Our findings offer an efficient, scalable path for building strong graph reasoners by finetuning LLMs with RL on graph-theoretic tasks, which combines the strengths of pretrained LLM capabilities with abundant, automatically generated synthetic data, suggesting that LLMs possess graph understanding abilities that RL can elicit successfully.
Autonomy via agents using large language models (LLMs) for personalized, standardized tasks boosts human efficiency. Automating web tasks (like booking hotels within a budget) is increasingly sought after. Fulfilling practical needs, the web agent also serves as an important proof-of-concept example for various agent grounding scenarios, with its success promising advancements in many future applications. Prior research often handcrafts web agent strategies (e.g., prompting templates, multi-agent systems, search methods, etc.) and the corresponding in-context examples, which may not generalize well across all real-world scenarios. On the other hand, there has been limited study on the misalignment between a web agent's observation/action representation and the pre-training data of the LLM it's based on. This discrepancy is especially notable when LLMs are primarily trained for language completion rather than tasks involving embodied navigation actions and symbolic web elements. Our study enhances an LLM-based web agent by simply refining its observation and action space to better align with the LLM's capabilities. This approach enables our base agent to significantly outperform previous methods on a wide variety of web tasks. Specifically, on WebArena, a benchmark featuring general-purpose web interaction tasks, our agent AgentOccam surpasses the previous state-of-the-art and concurrent work by 9.8 (+29.4%) and 5.9 (+15.8%) absolute points respectively, and boosts the success rate by 26.6 points (+161%) over similar plain web agents with its observation and action space alignment. We achieve this without using in-context examples, new agent roles, online feedback or search strategies. AgentOccam's simple design highlights LLMs' impressive zero-shot performance on web tasks, and underlines the critical role of carefully tuning observation and action spaces for LLM-based agents.
While Large Language Models have shown promise in cybersecurity applications, their effectiveness in identifying security threats within cloud deployments remains unexplored. This paper introduces AWS Cloud Security Engineering Eval, a novel dataset for evaluating LLMs cloud security threat modeling capabilities. ACSE-Eval contains 100 production grade AWS deployment scenarios, each featuring detailed architectural specifications, Infrastructure as Code implementations, documented security vulnerabilities, and associated threat modeling parameters. Our dataset enables systemic assessment of LLMs abilities to identify security risks, analyze attack vectors, and propose mitigation strategies in cloud environments. Our evaluations on ACSE-Eval demonstrate that GPT 4.1 and Gemini 2.5 Pro excel at threat identification, with Gemini 2.5 Pro performing optimally in 0-shot scenarios and GPT 4.1 showing superior results in few-shot settings. While GPT 4.1 maintains a slight overall performance advantage, Claude 3.7 Sonnet generates the most semantically sophisticated threat models but struggles with threat categorization and generalization. To promote reproducibility and advance research in automated cybersecurity threat analysis, we open-source our dataset, evaluation metrics, and methodologies.
Modern large language model (LLM) services increasingly rely on complex, often abstract operations, such as multi-step reasoning and multi-agent collaboration, to generate high-quality outputs. While users are billed based on token consumption and API usage, these internal steps are typically not visible. We refer to such systems as Commercial Opaque LLM Services (COLS). This position paper highlights emerging accountability challenges in COLS: users are billed for operations they cannot observe, verify, or contest. We formalize two key risks: \textit{quantity inflation}, where token and call counts may be artificially inflated, and \textit{quality downgrade}, where providers might quietly substitute lower-cost models or tools. Addressing these risks requires a diverse set of auditing strategies, including commitment-based, predictive, behavioral, and signature-based methods. We further explore the potential of complementary mechanisms such as watermarking and trusted execution environments to enhance verifiability without compromising provider confidentiality. We also propose a modular three-layer auditing framework for COLS and users that enables trustworthy verification across execution, secure logging, and user-facing auditability without exposing proprietary internals. Our aim is to encourage further research and policy development toward transparency, auditability, and accountability in commercial LLM services.
The integration of large language model (LLM) and data management (DATA) is rapidly redefining both domains. In this survey, we comprehensively review the bidirectional relationships. On the one hand, DATA4LLM, spanning large-scale data processing, storage, and serving, feeds LLMs with high quality, diversity, and timeliness of data required for stages like pre-training, post-training, retrieval-augmented generation, and agentic workflows: (i) Data processing for LLMs includes scalable acquisition, deduplication, filtering, selection, domain mixing, and synthetic augmentation; (ii) Data Storage for LLMs focuses on efficient data and model formats, distributed and heterogeneous storage hierarchies, KV-cache management, and fault-tolerant checkpointing; (iii) Data serving for LLMs tackles challenges in RAG (e.g., knowledge post-processing), LLM inference (e.g., prompt compression, data provenance), and training strategies (e.g., data packing and shuffling). On the other hand, in LLM4DATA, LLMs are emerging as general-purpose engines for data management. We review recent advances in (i) data manipulation, including automatic data cleaning, integration, discovery; (ii) data analysis, covering reasoning over structured, semi-structured, and unstructured data, and (iii) system optimization (e.g., configuration tuning, query rewriting, anomaly diagnosis), powered by LLM techniques like retrieval-augmented prompting, task-specialized fine-tuning, and multi-agent collaboration.
There has been growing interest in Multimodal Aspect-Based Sentiment Analysis (MABSA) in recent years. Existing methods predominantly rely on pre-trained small language models (SLMs) to collect information related to aspects and sentiments from both image and text, with an aim to align these two modalities. However, small SLMs possess limited capacity and knowledge, often resulting in inaccurate identification of meaning, aspects, sentiments, and their interconnections in textual and visual data. On the other hand, Large language models (LLMs) have shown exceptional capabilities in various tasks by effectively exploring fine-grained information in multimodal data. However, some studies indicate that LLMs still fall short compared to fine-tuned small models in the field of ABSA. Based on these findings, we propose a novel framework, termed LRSA, which combines the decision-making capabilities of SLMs with additional information provided by LLMs for MABSA. Specifically, we inject explanations generated by LLMs as rationales into SLMs and employ a dual cross-attention mechanism for enhancing feature interaction and fusion, thereby augmenting the SLMs' ability to identify aspects and sentiments. We evaluated our method using two baseline models, numerous experiments highlight the superiority of our approach on three widely-used benchmarks, indicating its generalizability and applicability to most pre-trained models for MABSA.