llm - 2025_06
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
Papers
We present a novel approach to the automated semantic analysis of legal texts using large language models (LLMs), targeting their transformation into formal representations in Defeasible Deontic Logic (DDL). We propose a structured pipeline that segments complex normative language into atomic snippets, extracts deontic rules, and evaluates them for syntactic and semantic coherence. Our methodology is evaluated across various LLM configurations, including prompt engineering strategies, fine-tuned models, and multi-stage pipelines, focusing on legal norms from the Australian Telecommunications Consumer Protections Code. Empirical results demonstrate promising alignment between machine-generated and expert-crafted formalizations, showing that LLMs - particularly when prompted effectively - can significantly contribute to scalable legal informatics.
Recent innovations in artificial intelligence (AI), primarily powered by large language models (LLMs), have transformed how programmers develop and maintain software -- leading to new frontiers in software engineering (SE). The advanced capabilities of LLM-based programming assistants to support software development tasks have led to a rise in the adoption of LLMs in SE. However, little is known about the evidenced-based practices, tools and processes verified by research findings, supported and adopted by AI programming assistants. To this end, our work conducts a preliminary evaluation exploring the beliefs of LLM used to support software development tasks. We investigate 17 evidence-based claims posited by empirical SE research across five LLM-based programming assistants. Our findings show that LLM-based programming assistants have ambiguous beliefs regarding research claims and lack credible evidence to support responses. Based on our results, we provide implications for practitioners adopting LLM-based programming assistants in development contexts and shed light on future research directions to enhance the reliability and trustworthiness of LLMs -- aiming to increase awareness and adoption of evidence-based SE research findings in practice.
Low-Rank Adaptation (LoRA) has emerged as a highly efficient framework for finetuning the weights of large foundation models, and has become the go-to method for data-driven customization of LLMs. Despite the promise of highly customized behaviors and capabilities, switching between relevant LoRAs in a multiturn setting is inefficient, as the key-value (KV) cache of the entire turn history must be recomputed with the LoRA weights before generation can begin. To address this problem, we propose Activated LoRA (aLoRA), an adapter architecture which modifies the LoRA framework to only adapt weights for the tokens in the sequence \emph{after} the aLoRA is invoked. This change crucially allows aLoRA to accept the base model's KV cache of the input string, meaning that aLoRA can be instantly activated whenever needed in a chain without recomputing the cache. This enables building what we call \emph{intrinsics}, i.e. specialized models invoked to perform well-defined operations on portions of an input chain or conversation that otherwise uses the base model by default. We train a set of aLoRA-based intrinsics models, demonstrating competitive accuracy with standard LoRA while achieving significant inference benefits. The codebase is at https://github.com/IBM/activated-lora.
Video large language models have not yet been widely deployed, largely due to their tendency to hallucinate. Typical benchmarks for Video-LLMs rely simply on multiple-choice questions. Unfortunately, VideoLLMs hallucinate far more aggressively on freeform text generation tasks like video captioning than they do on multiple choice verification tasks. To address this weakness, we propose ARGUS, a VideoLLM benchmark that measures freeform video captioning performance. By comparing VideoLLM outputs to human ground truth captions, ARGUS quantifies dual metrics. First, we measure the rate of hallucinations in the form of incorrect statements about video content or temporal relationships. Second, we measure the rate at which the model omits important descriptive details. Together, these dual metrics form a comprehensive view of video captioning performance.
Time series forecasting (TSF) is a fundamental and widely studied task, spanning methods from classical statistical approaches to modern deep learning and multimodal language modeling. Despite their effectiveness, these methods often follow a fast thinking paradigm emphasizing pattern extraction and direct value mapping, while overlooking explicit reasoning over temporal dynamics and contextual dependencies. Meanwhile, emerging slow-thinking LLMs (e.g., ChatGPT-o1, DeepSeek-R1) have demonstrated impressive multi-step reasoning capabilities across diverse domains, suggesting a new opportunity for reframing TSF as a structured reasoning task. This motivates a key question: can slow-thinking LLMs effectively reason over temporal patterns to support time series forecasting, even in zero-shot manner? To investigate this, in this paper, we propose TimeReasoner, an extensive empirical study that formulates TSF as a conditional reasoning task. We design a series of prompting strategies to elicit inference-time reasoning from pretrained slow-thinking LLMs and evaluate their performance across diverse TSF benchmarks. Our findings reveal that slow-thinking LLMs exhibit non-trivial zero-shot forecasting capabilities, especially in capturing high-level trends and contextual shifts. While preliminary, our study surfaces important insights into the reasoning behaviors of LLMs in temporal domains highlighting both their potential and limitations. We hope this work catalyzes further research into reasoning-based forecasting paradigms and paves the way toward more interpretable and generalizable TSF frameworks.
Financial analysis presents complex challenges that could leverage large language model (LLM) capabilities. However, the scarcity of challenging financial datasets, particularly for Japanese financial data, impedes academic innovation in financial analytics. As LLMs advance, this lack of accessible research resources increasingly hinders their development and evaluation in this specialized domain. To address this gap, we introduce EDINET-Bench, an open-source Japanese financial benchmark designed to evaluate the performance of LLMs on challenging financial tasks including accounting fraud detection, earnings forecasting, and industry prediction. EDINET-Bench is constructed by downloading annual reports from the past 10 years from Japan's Electronic Disclosure for Investors' NETwork (EDINET) and automatically assigning labels corresponding to each evaluation task. Our experiments reveal that even state-of-the-art LLMs struggle, performing only slightly better than logistic regression in binary classification for fraud detection and earnings forecasting. These results highlight significant challenges in applying LLMs to real-world financial applications and underscore the need for domain-specific adaptation. Our dataset, benchmark construction code, and evaluation code is publicly available to facilitate future research in finance with LLMs.
Query rewriting (QR) is a critical technique in e-commerce search, addressing the lexical gap between user queries and product descriptions to enhance search performance. Existing QR approaches typically fall into two categories: discriminative models and generative methods leveraging large language models (LLMs). Discriminative models often struggle with natural language understanding and offer limited flexibility in rewriting, while generative LLMs, despite producing high-quality rewrites, face high inference latency and cost in online settings. These limitations force offline deployment, making them vulnerable to issues like information staleness and semantic drift. To overcome these challenges, we propose a novel hybrid pipeline for QR that balances efficiency and effectiveness. Our approach combines offline knowledge distillation to create a lightweight but efficient student model with online reinforcement learning (RL) to refine query rewriting dynamically using real-time feedback. A key innovation is the use of LLMs as simulated human feedback, enabling scalable reward signals and cost-effective evaluation without manual annotations. Experimental results on Amazon ESCI dataset demonstrate significant improvements in query relevance, diversity, and adaptability, as well as positive feedback from the LLM simulation. This work contributes to advancing LLM capabilities for domain-specific applications, offering a robust solution for dynamic and complex e-commerce search environments.
Retrieving operational data from nuclear power plants requires exceptional accuracy and transparency due to the criticality of the decisions it supports. Traditionally, natural language to SQL (NL-to-SQL) approaches have been explored for querying such data. While NL-to-SQL promises ease of use, it poses significant risks: end-users cannot easily validate generated SQL queries, and legacy nuclear plant databases -- often complex and poorly structured -- complicate query generation due to decades of incremental modifications. These challenges increase the likelihood of inaccuracies and reduce trust in the approach. In this work, we propose an alternative paradigm: leveraging function-calling large language models (LLMs) to address these challenges. Instead of directly generating SQL queries, we define a set of pre-approved, purpose-specific functions representing common use cases. Queries are processed by invoking these functions, which encapsulate validated SQL logic. This hybrid approach mitigates the risks associated with direct NL-to-SQL translations by ensuring that SQL queries are reviewed and optimized by experts before deployment. While this strategy introduces the upfront cost of developing and maintaining the function library, we demonstrate how NL-to-SQL tools can assist in the initial generation of function code, allowing experts to focus on validation rather than creation. Our study includes a performance comparison between direct NL-to-SQL generation and the proposed function-based approach, highlighting improvements in accuracy and maintainability. This work underscores the importance of balancing user accessibility with operational safety and provides a novel, actionable framework for robust data retrieval in critical systems.
This study explores the application of in-context learning (ICL) to the dialogue state tracking (DST) problem and investigates the factors that influence its effectiveness. We use a sentence embedding based k-nearest neighbour method to retrieve the suitable demonstrations for ICL. The selected demonstrations, along with the test samples, are structured within a template as input to the LLM. We then conduct a systematic study to analyse the impact of factors related to demonstration selection and prompt context on DST performance. This work is conducted using the MultiWoZ2.4 dataset and focuses primarily on the OLMo-7B-instruct, Mistral-7B-Instruct-v0.3, and Llama3.2-3B-Instruct models. Our findings provide several useful insights on in-context learning abilities of LLMs for dialogue state tracking.
Recent advances of Reinforcement Learning (RL) have highlighted its potential in complex reasoning tasks, yet effective training often relies on external supervision, which limits the broader applicability. In this work, we propose a novel self-rewarding reinforcement learning framework to enhance Large Language Model (LLM) reasoning by leveraging the consistency of intermediate reasoning states across different reasoning trajectories. Our key insight is that correct responses often exhibit consistent trajectory patterns in terms of model likelihood: their intermediate reasoning states tend to converge toward their own final answers (high consistency) with minimal deviation toward other candidates (low volatility). Inspired by this observation, we introduce CoVo, an intrinsic reward mechanism that integrates Consistency and Volatility via a robust vector-space aggregation strategy, complemented by a curiosity bonus to promote diverse exploration. CoVo enables LLMs to perform RL in a self-rewarding manner, offering a scalable pathway for learning to reason without external supervision. Extensive experiments on diverse reasoning benchmarks show that CoVo achieves performance comparable to or even surpassing supervised RL. Our code is available at https://github.com/sastpg/CoVo.
While Generative AI stands to be one of the fastest adopted technologies ever, studies have made evident that the usage of Large Language Models (LLMs) puts significant burden on energy grids and our environment. It may prove a hindrance to the Sustainability goals of any organization. A crucial step in any Sustainability strategy is monitoring or estimating the energy consumption of various components. While there exist multiple tools for monitoring energy consumption, there is a dearth of tools/frameworks for estimating the consumption or carbon emissions. Current drawbacks of both monitoring and estimation tools include high input data points, intrusive nature, high error margin, etc. We posit that leveraging emerging LLM benchmarks and related data points can help overcome aforementioned challenges while balancing accuracy of the emission estimations. To that extent, we discuss the challenges of current approaches and present our evolving framework, R-ICE, which estimates prompt level inference carbon emissions by leveraging existing state-of-the-art(SOTA) benchmark. This direction provides a more practical and non-intrusive way to enable emerging use-cases like dynamic LLM routing, carbon accounting, etc. Our promising validation results suggest that benchmark-based modelling holds great potential for inference emission estimation and warrants further exploration from the scientific community.
Large language models (LLMs) have shown impressive capabilities on numerous natural language processing tasks. However, LLMs still struggle with numerical question answering for financial documents that include tabular and textual data. Recent works have showed the effectiveness of critic agents (i.e., self-correction) for this task given oracle labels. Building upon this framework, this paper examines the effectiveness of the traditional critic agent when oracle labels are not available, and show, through experiments, that this critic agent's performance deteriorates in this scenario. With this in mind, we present an improved critic agent, along with the calculator agent which outperforms the previous state-of-the-art approach (program-of-thought) and is safer. Furthermore, we investigate how our agents interact with each other, and how this interaction affects their performance.
Large Language Models (LLMs) have rapidly evolved over the past few years and are currently evaluated for their efficacy within the domain of offensive cyber-security. While initial forays showcase the potential of LLMs to enhance security research, they also raise critical ethical concerns regarding the dual-use of offensive security tooling. This paper analyzes a set of papers that leverage LLMs for offensive security, focusing on how ethical considerations are expressed and justified in their work. The goal is to assess the culture of AI in offensive security research regarding ethics communication, highlighting trends, best practices, and gaps in current discourse. We provide insights into how the academic community navigates the fine line between innovation and ethical responsibility. Particularly, our results show that 13 of 15 reviewed prototypes (86.6\%) mentioned ethical considerations and are thus aware of the potential dual-use of their research. Main motivation given for the research was allowing broader access to penetration-testing as well as preparing defenders for AI-guided attackers.
A significant portion of the energy consumed by Large Language Models (LLMs) arises from their inference processes; hence developing energy-efficient methods for inference is crucial. While several techniques exist for inference optimization, output compression remains relatively unexplored, with only a few preliminary efforts addressing this aspect. In this work, we first benchmark 12 decoder-only LLMs across 5 datasets, revealing that these models often produce responses that are substantially longer than necessary. We then conduct a comprehensive quality assessment of LLM responses, formally defining six information categories present in LLM responses. We show that LLMs often tend to include redundant or additional information besides the minimal answer. To address this issue of long responses by LLMs, we explore several simple and intuitive prompt-engineering strategies. Empirical evaluation shows that appropriate prompts targeting length reduction and controlling information content can achieve significant energy optimization between 25-60\% by reducing the response length while preserving the quality of LLM responses.
Large Language Models (LLMs) have demonstrated strong capabilities in general-purpose code generation. However, generating the code which is deeply hardware-specific, architecture-aware, and performance-critical, especially for massively parallel GPUs, remains a complex challenge. In this work, we explore the use of LLMs for the automated generation and optimization of CUDA programs, with the goal of producing high-performance GPU kernels that fully exploit the underlying hardware. To address this challenge, we propose a novel framework called \textbf{Feature Search and Reinforcement (FSR)}. FSR jointly optimizes compilation and functional correctness, as well as the runtime performance, which are validated through extensive and diverse test cases, and measured by actual kernel execution latency on the target GPU, respectively. This approach enables LLMs not only to generate syntactically and semantically correct CUDA code but also to iteratively refine it for efficiency, tailored to the characteristics of the GPU architecture. We evaluate FSR on representative CUDA kernels, covering AI workloads and computational intensive algorithms. Our results show that LLMs augmented with FSR consistently guarantee correctness rates. Meanwhile, the automatically generated kernels can outperform general human-written code by a factor of up to 179$\times$ in execution speeds. These findings highlight the potential of combining LLMs with performance reinforcement to automate GPU programming for hardware-specific, architecture-sensitive, and performance-critical applications.
Optimal hyperparameter selection is critical for maximizing neural network performance, especially as models grow in complexity. This work investigates the viability of leveraging large language models (LLMs) for hyperparameter optimization by fine-tuning a parameter-efficient version of Code Llama using LoRA. The adapted LLM is capable of generating accurate and efficient hyperparameter recommendations tailored to diverse neural network architectures. Unlike traditional approaches such as Optuna, which rely on computationally intensive trial-and-error procedures, our method achieves competitive or superior results in terms of Root Mean Square Error (RMSE) while significantly reducing computational overhead. Our findings demonstrate that LLM-based optimization not only matches the performance of state-of-the-art techniques like Tree-structured Parzen Estimators (TPE) but also substantially accelerates the tuning process. This positions LLMs as a promising alternative for rapid experimentation, particularly in resource-constrained environments such as edge devices and mobile platforms, where computational efficiency is essential. In addition to improved efficiency, the method offers time savings and consistent performance across various tasks, highlighting its robustness and generalizability. All generated hyperparameters are included in the LEMUR Neural Network (NN) Dataset, which is publicly available and serves as an open-source benchmark for hyperparameter optimization research.
In this work, we approach spoken Dialogue State Tracking (DST) by bridging the representation spaces of speech encoders and LLMs via a small connector module, with a focus on fully open-sourced and open-data components (WavLM-large, OLMo). We focus on ablating different aspects of such systems including full/LoRA adapter fine-tuning, the effect of agent turns in the dialogue history, as well as fuzzy matching-based output post-processing, which greatly improves performance of our systems on named entities in the dialogue slot values. We conduct our experiments on the SpokenWOZ dataset, and additionally utilize the Speech-Aware MultiWOZ dataset to augment our training data. Ultimately, our best-performing WavLM + connector + OLMo-1B aligned models achieve state of the art on the SpokenWOZ test set (34.66% JGA), and our system with Gemma-2-9B-instruct further surpasses this result, reaching 42.17% JGA on SpokenWOZ test.
The requirements engineering (RE) phase is pivotal in developing high-quality software. Integrating advanced modelling techniques with large language models (LLMs) and formal verification in a logical style can significantly enhance this process. We propose a comprehensive framework that focuses on specific Unified Modelling Language (UML) diagrams for preliminary system development. This framework offers visualisations at various modelling stages and seamlessly integrates large language models and logical reasoning engines. The behavioural models generated with the assistance of LLMs are automatically translated into formal logical specifications. Deductive formal verification ensures that logical requirements and interrelations between software artefacts are thoroughly addressed. Ultimately, the framework facilitates the automatic generation of program skeletons, streamlining the transition from design to implementation.
Feature transformation enhances data representation by deriving new features from the original data. Generative AI offers potential for this task, but faces challenges in stable generation (consistent outputs) and valid generation (error-free sequences). Existing methods--traditional MLs' low validity and LLMs' instability--fail to resolve both. We find that LLMs ensure valid syntax, while ML's gradient-steered search stabilizes performance. To bridge this gap, we propose a teaming framework combining LLMs' symbolic generation with ML's gradient optimization. This framework includes four steps: (1) golden examples generation, aiming to prepare high-quality samples with the ground knowledge of the teacher LLM; (2) feature transformation sequence embedding and search, intending to uncover potentially superior embeddings within the latent space; (3) student LLM feature transformation, aiming to distill knowledge from the teacher LLM; (4) LLM-ML decoder teaming, dedicating to combine ML and the student LLM probabilities for valid and stable generation. The experiments on various datasets show that the teaming policy can achieve 5\% improvement in downstream performance while reducing nearly half of the error cases. The results also demonstrate the efficiency and robustness of the teaming policy. Additionally, we also have exciting findings on LLMs' capacity to understand the original data.
The rapid growth of the low-altitude economy has driven the widespread adoption of unmanned aerial vehicles (UAVs). This growing deployment presents new challenges for UAV trajectory planning in complex urban environments. However, existing studies often overlook key factors, such as urban airspace constraints and economic efficiency, which are essential in low-altitude economy contexts. Deep reinforcement learning (DRL) is regarded as a promising solution to these issues, while its practical adoption remains limited by low learning efficiency. To overcome this limitation, we propose a novel UAV trajectory planning framework that combines DRL with large language model (LLM) reasoning to enable safe, compliant, and economically viable path planning. Experimental results demonstrate that our method significantly outperforms existing baselines across multiple metrics, including data collection rate, collision avoidance, successful landing, regulatory compliance, and energy efficiency. These results validate the effectiveness of our approach in addressing UAV trajectory planning key challenges under constraints of the low-altitude economy networking.
The growing use of large language model (LLM)-based chatbots has raised concerns about fairness. Fairness issues in LLMs can lead to severe consequences, such as bias amplification, discrimination, and harm to marginalized communities. While existing fairness benchmarks mainly focus on single-turn dialogues, multi-turn scenarios, which in fact better reflect real-world conversations, present greater challenges due to conversational complexity and potential bias accumulation. In this paper, we propose a comprehensive fairness benchmark for LLMs in multi-turn dialogue scenarios, \textbf{FairMT-Bench}. Specifically, we formulate a task taxonomy targeting LLM fairness capabilities across three stages: context understanding, user interaction, and instruction trade-offs, with each stage comprising two tasks. To ensure coverage of diverse bias types and attributes, we draw from existing fairness datasets and employ our template to construct a multi-turn dialogue dataset, \texttt{FairMT-10K}. For evaluation, GPT-4 is applied, alongside bias classifiers including Llama-Guard-3 and human validation to ensure robustness. Experiments and analyses on \texttt{FairMT-10K} reveal that in multi-turn dialogue scenarios, current LLMs are more likely to generate biased responses, and there is significant variation in performance across different tasks and models. Based on this, we curate a challenging dataset, \texttt{FairMT-1K}, and test 15 current state-of-the-art (SOTA) LLMs on this dataset. The results show the current state of fairness in LLMs and showcase the utility of this novel approach for assessing fairness in more realistic multi-turn dialogue contexts, calling for future work to focus on LLM fairness improvement and the adoption of \texttt{FairMT-1K} in such efforts.
When complex SQL queries suffer slow executions despite query optimization, DBAs typically invoke automated query rewriting tools to recommend ``lean'' equivalents that are conducive to faster execution. The rewritings are usually achieved via transformation rules, but these rules are limited in scope and difficult to update in a production system. Recently, LLM-based techniques have also been suggested, but they are prone to semantic and syntactic errors. We investigate here how the remarkable cognitive capabilities of LLMs can be leveraged for performant query rewriting while incorporating safeguards and optimizations to ensure correctness and efficiency. Our study shows that these goals can be progressively achieved through incorporation of (a) an ensemble suite of basic prompts, (b) database-sensitive prompts via redundancy removal and selectivity-based rewriting rules, and (c) LLM token probability-guided rewrite paths. Further, a suite of logic-based and statistical tools can be used to check for semantic violations in the rewrites prior to DBA consideration. We have implemented the above LLM-infused techniques in the LITHE system, and evaluated complex analytic queries from standard benchmarks on contemporary database platforms. The results show significant performance improvements for slow queries, with regard to both abstract costing and actual execution, over both SOTA techniques and the native query optimizer. For instance, with TPC-DS on PostgreSQL, the geometric mean of the runtime speedups for slow queries was as high as 13.2 over the native optimizer, whereas SOTA delivered 4.9 in comparison. Overall, LITHE is a promising step toward viable LLM-based advisory tools for ameliorating enterprise query performance.
Video Temporal Grounding (VTG), which aims to localize video clips corresponding to natural language queries, is a fundamental yet challenging task in video understanding. Existing Transformer-based methods often suffer from redundant attention and suboptimal multi-modal alignment. To address these limitations, we propose MLVTG, a novel framework that integrates two key modules: MambaAligner and LLMRefiner. MambaAligner uses stacked Vision Mamba blocks as a backbone instead of Transformers to model temporal dependencies and extract robust video representations for multi-modal alignment. LLMRefiner leverages the specific frozen layer of a pre-trained Large Language Model (LLM) to implicitly transfer semantic priors, enhancing multi-modal alignment without fine-tuning. This dual alignment strategy, temporal modeling via structured state-space dynamics and semantic purification via textual priors, enables more precise localization. Extensive experiments on QVHighlights, Charades-STA, and TVSum demonstrate that MLVTG achieves state-of-the-art performance and significantly outperforms existing baselines.
Retrieval Augmented Generation (RAG) is a commonly used approach for enhancing large language models (LLMs) with relevant and up-to-date information. However, the retrieved sources can often contain conflicting information and it remains unclear how models should address such discrepancies. In this work, we first propose a novel taxonomy of knowledge conflict types in RAG, along with the desired model behavior for each type. We then introduce CONFLICTS, a high-quality benchmark with expert annotations of conflict types in a realistic RAG setting. CONFLICTS is the first benchmark that enables tracking progress on how models address a wide range of knowledge conflicts. We conduct extensive experiments on this benchmark, showing that LLMs often struggle to appropriately resolve conflicts between sources. While prompting LLMs to explicitly reason about the potential conflict in the retrieved documents significantly improves the quality and appropriateness of their responses, substantial room for improvement in future research remains.
In recent years, researchers have started analyzing the cultural sensitivity of LLMs. In this respect, Etiquettes have been an active area of research. Etiquettes are region-specific and are an essential part of the culture of a region; hence, it is imperative to make LLMs sensitive to etiquettes. However, there needs to be more resources in evaluating LLMs for their understanding and bias with regard to etiquettes. In this resource paper, we introduce EtiCor++, a corpus of etiquettes worldwide. We introduce different tasks for evaluating LLMs for knowledge about etiquettes across various regions. Further, we introduce various metrics for measuring bias in LLMs. Extensive experimentation with LLMs shows inherent bias towards certain regions.
Which large language model (LLM) is better? Every evaluation tells a story, but what do users really think about current LLMs? This paper presents CLUE, an LLM-powered interviewer that conducts in-the-moment user experience interviews, right after users interact with LLMs, and automatically gathers insights about user opinions from massive interview logs. We conduct a study with thousands of users to understand user opinions on mainstream LLMs, recruiting users to first chat with a target LLM and then be interviewed by CLUE. Our experiments demonstrate that CLUE captures interesting user opinions, e.g., the bipolar views on the displayed reasoning process of DeepSeek-R1 and demands for information freshness and multi-modality. Our code and data are at https://github.com/cxcscmu/LLM-Interviewer.
Large Language Models $($LLMs$)$ solve complex problems using training-free methods like prompt engineering and in-context learning, yet ensuring reasoning correctness remains challenging. While self-correction methods such as self-consistency and self-refinement aim to improve reliability, they often reinforce biases due to the lack of effective feedback mechanisms. Multi-Agent Debate $($MAD$)$ has emerged as an alternative, but we identify two key limitations: bias reinforcement, where debate amplifies model biases instead of correcting them, and lack of perspective diversity, as all agents share the same model and reasoning patterns, limiting true debate effectiveness. To systematically evaluate these issues, we introduce $\textit{MetaNIM Arena}$, a benchmark designed to assess LLMs in adversarial strategic decision-making, where dynamic interactions influence optimal decisions. To overcome MAD's limitations, we propose $\textbf{DReaMAD}$ $($$\textbf{D}$iverse $\textbf{Rea}$soning via $\textbf{M}$ulti-$\textbf{A}$gent $\textbf{D}$ebate with Refined Prompt$)$, a novel framework that $(1)$ refines LLM's strategic prior knowledge to improve reasoning quality and $(2)$ promotes diverse viewpoints within a single model by systematically modifying prompts, reducing bias. Empirical results show that $\textbf{DReaMAD}$ significantly improves decision accuracy, reasoning diversity, and bias mitigation across multiple strategic tasks, establishing it as a more effective approach for LLM-based decision-making.
Large Language models (LLMs) have demonstrated impressive performance on a wide range of tasks, including in multimodal settings such as speech. However, their evaluation is often limited to English and a few high-resource languages. For low-resource languages, there is no standardized evaluation benchmark. In this paper, we address this gap by introducing mSTEB, a new benchmark to evaluate the performance of LLMs on a wide range of tasks covering language identification, text classification, question answering, and translation tasks on both speech and text modalities. We evaluated the performance of leading LLMs such as Gemini 2.0 Flash and GPT-4o (Audio) and state-of-the-art open models such as Qwen 2 Audio and Gemma 3 27B. Our evaluation shows a wide gap in performance between high-resource and low-resource languages, especially for languages spoken in Africa and Americas/Oceania. Our findings show that more investment is needed to address their under-representation in LLMs coverage.
Large Language Models (LLMs) have recently achieved remarkable progress by leveraging Reinforcement Learning and extended Chain-of-Thought (CoT) techniques. However, the challenge of performing efficient language reasoning--especially during inference with extremely long outputs--has drawn increasing attention from the research community. In this work, we propose a dynamic ratio-based training pipeline that does not rely on sophisticated data annotations or interpolation between multiple models. We continuously balance the weights between the model's System-1 and System-2 data to eliminate redundant reasoning processes while preserving the model's reasoning capability. We validate our approach across models on DeepSeek-R1-Distill-7B and DeepSeek-R1-Distill-14B and on a diverse set of benchmarks with varying difficulty levels. Our method significantly reduces the number of output tokens by nearly 40% while maintaining the accuracy of the reasoning. Our code and data will be available soon.
Leveraging more test-time computation has proven to be an effective way to boost the reasoning capabilities of large language models (LLMs). Among various methods, the verify-and-improve paradigm stands out for enabling dynamic solution exploration and feedback incorporation. However, existing approaches often suffer from restricted feedback spaces and lack of coordinated training of different parties, leading to suboptimal performance. To address this, we model this multi-turn refinement process as a Markov Decision Process and introduce DPSDP (Direct Policy Search by Dynamic Programming), a reinforcement learning algorithm that trains an actor-critic LLM system to iteratively refine answers via direct preference learning on self-generated data. Theoretically, DPSDP can match the performance of any policy within the training distribution. Empirically, we instantiate DPSDP with various base models and show improvements on both in- and out-of-distribution benchmarks. For example, on benchmark MATH 500, majority voting over five refinement steps increases first-turn accuracy from 58.2% to 63.2% with Ministral-based models. An ablation study further confirms the benefits of multi-agent collaboration and out-of-distribution generalization.
Optimizing inference for long-context Large Language Models (LLMs) is increasingly important due to the quadratic compute and linear memory complexity of Transformers. Existing approximation methods, such as key-value (KV) cache dropping, sparse attention, and prompt compression, typically rely on rough predictions of token or KV pair importance. We propose a novel framework for approximate LLM inference that leverages small draft models to more accurately predict the importance of tokens and KV pairs. Specifically, we introduce two instantiations of our proposed framework: (i) SpecKV, which leverages a draft output to accurately assess the importance of each KV pair for more effective KV cache dropping, and (ii) SpecPC, which uses the draft model's attention activations to identify and discard unimportant prompt tokens. To the best of our knowledge, this is the first work to use draft models for approximate LLM inference acceleration, extending their utility beyond traditional lossless speculative decoding. We motivate our methods with theoretical and empirical analyses, and show a strong correlation between the attention patterns of draft and target models. Extensive experiments on long-context benchmarks show that our methods consistently achieve higher accuracy than existing baselines, while preserving the same improvements in memory usage, latency, and throughput. Our code is available at https://github.com/furiosa-ai/draft-based-approx-llm.
Modeling urban crime is an important yet challenging task that requires understanding the subtle visual, social, and cultural cues embedded in urban environments. Previous work has mainly focused on rule-based agent-based modeling (ABM) and deep learning methods. ABMs offer interpretability of internal mechanisms but exhibit limited predictive accuracy. In contrast, deep learning methods are often effective in prediction but are less interpretable and require extensive training data. Moreover, both lines of work lack the cognitive flexibility to adapt to changing environments. Leveraging the capabilities of large language models (LLMs), we propose CrimeMind, a novel LLM-driven ABM framework for simulating urban crime within a multi-modal urban context. A key innovation of our design is the integration of the Routine Activity Theory (RAT) into the agentic workflow of CrimeMind, enabling it to process rich multi-modal urban features and reason about criminal behavior. However, RAT requires LLM agents to infer subtle cues in evaluating environmental safety as part of assessing guardianship, which can be challenging for LLMs. To address this, we collect a small-scale human-annotated dataset and align CrimeMind's perception with human judgment via a training-free textual gradient method. Experiments across four major U.S. cities demonstrate that CrimeMind outperforms both traditional ABMs and deep learning baselines in crime hotspot prediction and spatial distribution accuracy, achieving up to a 24% improvement over the strongest baseline. Furthermore, we conduct counterfactual simulations of external incidents and policy interventions and it successfully captures the expected changes in crime patterns, demonstrating its ability to reflect counterfactual scenarios. Overall, CrimeMind enables fine-grained modeling of individual behaviors and facilitates evaluation of real-world interventions.
Inference-time steering aims to alter the response characteristics of large language models (LLMs) without modifying their underlying parameters. A critical step in this process is the identification of internal modules within LLMs that are associated with the target behavior. However, current approaches to module selection often depend on superficial cues or ad-hoc heuristics, which can result in suboptimal or unintended outcomes. In this work, we propose a principled causal-attribution framework for identifying behavior-relevant attention heads in transformers. For each head, we train a vector-quantized autoencoder (VQ-AE) on its attention activations, partitioning the latent space into behavior-relevant and behavior-irrelevant subspaces, each quantized with a shared learnable codebook. We assess the behavioral relevance of each head by quantifying the separability of VQ-AE encodings for behavior-aligned versus behavior-violating responses using a binary classification metric. This yields a behavioral relevance score that reflects each head discriminative capacity with respect to the target behavior, guiding both selection and importance weighting. Experiments on seven LLMs from two model families and five behavioral steering datasets demonstrate that our method enables more accurate inference-time interventions, achieving superior performance on the truthfulness-steering task. Furthermore, the heads selected by our approach exhibit strong zero-shot generalization in cross-domain truthfulness-steering scenarios.
Large language models (LLMs) have demonstrated remarkable performance on various medical benchmarks, but their capabilities across different cognitive levels remain underexplored. Inspired by Bloom's Taxonomy, we propose a multi-cognitive-level evaluation framework for assessing LLMs in the medical domain in this study. The framework integrates existing medical datasets and introduces tasks targeting three cognitive levels: preliminary knowledge grasp, comprehensive knowledge application, and scenario-based problem solving. Using this framework, we systematically evaluate state-of-the-art general and medical LLMs from six prominent families: Llama, Qwen, Gemma, Phi, GPT, and DeepSeek. Our findings reveal a significant performance decline as cognitive complexity increases across evaluated models, with model size playing a more critical role in performance at higher cognitive levels. Our study highlights the need to enhance LLMs' medical capabilities at higher cognitive levels and provides insights for developing LLMs suited to real-world medical applications.
In-situ LLM inference on end-user devices has gained significant interest due to its privacy benefits and reduced dependency on external infrastructure. However, as the decoding process is memory-bandwidth-bound, the diverse processing units in modern end-user devices cannot be fully exploited, resulting in slow LLM inference. This paper presents Ghidorah, a LLM inference system for end-user devices with the unified memory architecture. The key idea of Ghidorah can be summarized in two steps: 1) leveraging speculative decoding approaches to enhance parallelism, and 2) ingeniously distributing workloads across multiple heterogeneous processing units to maximize computing power utilization. Ghidorah includes the hetero-core model parallelism (HCMP) architecture and the architecture-aware profiling (ARCA) approach. The HCMP architecture guides partitioning by leveraging the unified memory design of end-user devices and adapting to the hybrid computational demands of speculative decoding. The ARCA approach is used to determine the optimal speculative strategy and partitioning strategy, balancing acceptance rate with parallel capability to maximize the speedup. Additionally, we optimize sparse computation on ARM CPUs. Experimental results show that Ghidorah can achieve up to 7.6x speedup in the dominant LLM decoding phase compared to the sequential decoding approach in NVIDIA Jetson NX.
Large Language Models (LLMs) excel in traditional natural language processing tasks but struggle with problems that require complex domain-specific calculations or simulations. While equipping LLMs with external tools to build LLM-based agents can enhance their capabilities, existing approaches lack the flexibility to address diverse and ever-evolving user queries in open domains. Currently, there is also no existing dataset that evaluates LLMs on open-domain knowledge that requires tools to solve. To this end, we introduce OpenAct benchmark to evaluate the open-domain task-solving capability, which is built on human expert consultation and repositories in GitHub. It comprises 339 questions spanning 7 diverse domains that need to be solved with domain-specific methods. In our experiments, even state-of-the-art LLMs and LLM-based agents demonstrate unsatisfactory success rates, underscoring the need for a novel approach. Furthermore, we present OpenAgent, a novel LLM-based agent system that can tackle evolving queries in open domains through autonomously integrating specialized tools from GitHub. OpenAgent employs 1) a hierarchical framework where specialized agents handle specific tasks and can assign tasks to inferior agents, 2) a bi-level experience learning mechanism to learn from both humans' and its own experiences to tackle tool flaws. Experiments demonstrate its superior effectiveness and efficiency, which significantly outperforms baselines. Our data and code are open-source at https://github.com/OpenBMB/OpenAct.
Generative AI technologies, particularly Large Language Models (LLMs), are rapidly being adopted across industry, academia, and government sectors, owing to their remarkable capabilities in natural language processing. However, despite their strengths, the inconsistency and unpredictability of LLM outputs present substantial challenges, especially in security-critical domains such as access control. One critical issue that emerges prominently is the consistency of LLM-generated responses, which is paramount for ensuring secure and reliable operations. In this paper, we study the application of LLMs within the context of Cybersecurity Access Control Systems. Specifically, we investigate the consistency and accuracy of LLM-generated password policies, translating natural language prompts into executable pwquality.conf configuration files. Our experimental methodology adopts two distinct approaches: firstly, we utilize pre-trained LLMs to generate configuration files purely from natural language prompts without additional guidance. Secondly, we provide these models with official pwquality.conf documentation to serve as an informative baseline. We systematically assess the soundness, accuracy, and consistency of these AI-generated configurations. Our findings underscore significant challenges in the current generation of LLMs and contribute valuable insights into refining the deployment of LLMs in Access Control Systems.
Machine learning is advancing rapidly, with applications bringing notable benefits, such as improvements in translation and code generation. Models like ChatGPT, powered by Large Language Models (LLMs), are increasingly integrated into daily life. However, alongside these benefits, LLMs also introduce social risks. Malicious users can exploit LLMs by submitting harmful prompts, such as requesting instructions for illegal activities. To mitigate this, models often include a security mechanism that automatically rejects such harmful prompts. However, they can be bypassed through LLM jailbreaks. Current jailbreaks often require significant manual effort, high computational costs, or result in excessive model modifications that may degrade regular utility. We introduce TwinBreak, an innovative safety alignment removal method. Building on the idea that the safety mechanism operates like an embedded backdoor, TwinBreak identifies and prunes parameters responsible for this functionality. By focusing on the most relevant model layers, TwinBreak performs fine-grained analysis of parameters essential to model utility and safety. TwinBreak is the first method to analyze intermediate outputs from prompts with high structural and content similarity to isolate safety parameters. We present the TwinPrompt dataset containing 100 such twin prompts. Experiments confirm TwinBreak's effectiveness, achieving 89% to 98% success rates with minimal computational requirements across 16 LLMs from five vendors.
Test smells indicate poor development practices in test code, reducing maintainability and reliability. While developers often struggle to prevent or refactor these issues, existing tools focus primarily on detection rather than automated refactoring. Large Language Models (LLMs) have shown strong potential in code understanding and transformation, but their ability to both identify and refactor test smells remains underexplored. We evaluated GPT-4-Turbo, LLaMA 3 70B, and Gemini-1.5 Pro on Python and Java test suites, using PyNose and TsDetect for initial smell detection, followed by LLM-driven refactoring. Gemini achieved the highest detection accuracy (74.35\% Python, 80.32\% Java), while LLaMA was lowest. All models could refactor smells, but effectiveness varied, sometimes introducing new smells. Gemini also improved test coverage, unlike GPT-4 and LLaMA, which often reduced it. These results highlight LLMs' potential for automated test smell refactoring, with Gemini as the strongest performer, though challenges remain across languages and smell types.
Automatic indoor layout generation has attracted increasing attention due to its potential in interior design, virtual environment construction, and embodied AI. Existing methods fall into two categories: prompt-driven approaches that leverage proprietary LLM services (e.g., GPT APIs) and learning-based methods trained on layout data upon diffusion-based models. Prompt-driven methods often suffer from spatial inconsistency and high computational costs, while learning-based methods are typically constrained by coarse relational graphs and limited datasets, restricting their generalization to diverse room categories. In this paper, we revisit LLM-based indoor layout generation and present 3D-SynthPlace, a large-scale dataset that combines synthetic layouts generated via a 'GPT synthesize, Human inspect' pipeline, upgraded from the 3D-Front dataset. 3D-SynthPlace contains nearly 17,000 scenes, covering four common room types -- bedroom, living room, kitchen, and bathroom -- enriched with diverse objects and high-level spatial annotations. We further introduce OptiScene, a strong open-source LLM optimized for indoor layout generation, fine-tuned based on our 3D-SynthPlace dataset through our two-stage training. For the warum-up stage I, we adopt supervised fine-tuning (SFT), which is taught to first generate high-level spatial descriptions then conditionally predict concrete object placements. For the reinforcing stage II, to better align the generated layouts with human design preferences, we apply multi-turn direct preference optimization (DPO), which significantly improving layout quality and generation success rates. Extensive experiments demonstrate that OptiScene outperforms traditional prompt-driven and learning-based baselines. Moreover, OptiScene shows promising potential in interactive tasks such as scene editing and robot navigation.
Driven by advances in Large Language Models (LLMs), integrating them into recommendation tasks has gained interest due to their strong semantic understanding and prompt flexibility. Prior work encoded user-item interactions or metadata into prompts for recommendations. In parallel, LLM reasoning, boosted by test-time scaling and reinforcement learning, has excelled in fields like mathematics and code, where reasoning traces and correctness signals are clear, enabling high performance and interpretability. However, directly applying these reasoning methods to recommendation is ineffective because user feedback is implicit and lacks reasoning supervision. To address this, we propose $\textbf{R2Rec}$, a reasoning-enhanced recommendation framework that samples interaction chains from the user-item graph and converts them into structured interaction-of-thoughts via a progressive masked prompting strategy, with each thought representing stepwise reasoning grounded in interaction context. This allows LLMs to simulate step-by-step decision-making based on implicit patterns. We design a two-stage training pipeline: supervised fine-tuning teaches basic reasoning from high-quality traces, and reinforcement learning refines reasoning via reward signals, alleviating sparse explicit supervision. Experiments on three real-world datasets show R2Rec outperforms classical and LLM-based baselines with an average $\textbf{10.48%}$ improvement in HitRatio@1 and $\textbf{131.81%}$ gain over the original LLM. Furthermore, the explicit reasoning chains enhance interpretability by revealing the decision process. Our code is available at: https://anonymous.4open.science/r/R2Rec-7C5D.
While Large Language Models (LLMs) have achieved remarkable success in a wide range of applications, their performance often degrades in complex reasoning tasks. In this work, we introduce SELT (Self-Evaluation LLM Tree Search), a novel framework that leverages a modified Monte Carlo Tree Search (MCTS) to enhance LLM reasoning without relying on external reward models. By redefining the Upper Confidence Bound scoring to align with intrinsic self-evaluation capabilities of LLMs and decomposing the inference process into atomic subtasks augmented with semantic clustering at each node, SELT effectively balances exploration and exploitation, reduces redundant reasoning paths, and mitigates hallucination. We validate our approach on challenging benchmarks, including the knowledge-based MMLU and the Tool Learning dataset Seal-Tools, where SELT achieves significant improvements in answer accuracy and reasoning robustness compared to baseline methods. Notably, our framework operates without task-specific fine-tuning, demonstrating strong generalizability across diverse reasoning tasks. Relevant results and code are available at https://github.com/fairyshine/SELT .
Large language models (LLMs) have recently demonstrated promising capabilities in chemistry tasks while still facing challenges due to outdated pretraining knowledge and the difficulty of incorporating specialized chemical expertise. To address these issues, we propose an LLM-based agent that synergistically integrates 137 external chemical tools created ranging from basic information retrieval to complex reaction predictions, and a dataset curation pipeline to generate the dataset ChemToolBench that facilitates both effective tool selection and precise parameter filling during fine-tuning and evaluation. We introduce a Hierarchical Evolutionary Monte Carlo Tree Search (HE-MCTS) framework, enabling independent optimization of tool planning and execution. By leveraging self-generated data, our approach supports step-level fine-tuning (FT) of the policy model and training task-adaptive PRM and ORM that surpass GPT-4o. Experimental evaluations demonstrate that our approach significantly improves performance in Chemistry QA and discovery tasks, offering a robust solution to integrate specialized tools with LLMs for advanced chemical applications. All datasets and code are available at https://github.com/AI4Chem/ChemistryAgent .
One of the primary challenges in optimizing large language models (LLMs) for long-context inference lies in the high memory consumption of the Key-Value (KV) cache. Existing approaches, such as quantization, have demonstrated promising results in reducing memory usage. However, current quantization methods cannot take both effectiveness and efficiency into account. In this paper, we propose MoQAE, a novel mixed-precision quantization method via mixture of quantization-aware experts. First, we view different quantization bit-width configurations as experts and use the traditional mixture of experts (MoE) method to select the optimal configuration. To avoid the inefficiency caused by inputting tokens one by one into the router in the traditional MoE method, we input the tokens into the router chunk by chunk. Second, we design a lightweight router-only fine-tuning process to train MoQAE with a comprehensive loss to learn the trade-off between model accuracy and memory usage. Finally, we introduce a routing freezing (RF) and a routing sharing (RS) mechanism to further reduce the inference overhead. Extensive experiments on multiple benchmark datasets demonstrate that our method outperforms state-of-the-art KV cache quantization approaches in both efficiency and effectiveness.
LLM agents are increasingly deployed to automate real-world tasks by invoking APIs through natural language instructions. While powerful, they often suffer from misinterpretation of user intent, leading to the agent's actions that diverge from the user's intended goal, especially as external toolkits evolve. Traditional software testing assumes structured inputs and thus falls short in handling the ambiguity of natural language. We introduce IntenTest, an API-centric stress testing framework that systematically uncovers intent integrity violations in LLM agents. Unlike prior work focused on fixed benchmarks or adversarial inputs, IntenTest generates realistic tasks based on toolkits' documentation and applies targeted mutations to expose subtle agent errors while preserving user intent. To guide testing, we propose semantic partitioning, which organizes natural language tasks into meaningful categories based on toolkit API parameters and their equivalence classes. Within each partition, seed tasks are mutated and ranked by a lightweight predictor that estimates the likelihood of triggering agent errors. To enhance efficiency, IntenTest maintains a datatype-aware strategy memory that retrieves and adapts effective mutation patterns from past cases. Experiments on 80 toolkit APIs demonstrate that IntenTest effectively uncovers intent integrity violations, significantly outperforming baselines in both error-exposing rate and query efficiency. Moreover, IntenTest generalizes well to stronger target models using smaller LLMs for test generation, and adapts to evolving APIs across domains.
Tokenization methods like Byte-Pair Encoding (BPE) enhance computational efficiency in large language models (LLMs) but often obscure internal character structures within tokens. This limitation hinders LLMs' ability to predict precise character positions, which is crucial in tasks like Chinese Spelling Correction (CSC) where identifying the positions of misspelled characters accelerates correction processes. We propose Token Internal Position Awareness (TIPA), a method that significantly improves models' ability to capture character positions within tokens by training them on reverse character prediction tasks using the tokenizer's vocabulary. Experiments demonstrate that TIPA enhances position prediction accuracy in LLMs, enabling more precise identification of target characters in original text. Furthermore, when applied to downstream tasks that do not require exact position prediction, TIPA still boosts performance in tasks needing character-level information, validating its versatility and effectiveness.
GA LLM is a hybrid framework that combines Genetic Algorithms with Large Language Models to handle structured generation tasks under strict constraints. Each output, such as a plan or report, is treated as a gene, and evolutionary operations like selection, crossover, and mutation are guided by the language model to iteratively improve solutions. The language model provides domain knowledge and creative variation, while the genetic algorithm ensures structural integrity and global optimization. GA LLM has proven effective in tasks such as itinerary planning, academic outlining, and business reporting, consistently producing well structured and requirement satisfying results. Its modular design also makes it easy to adapt to new tasks. Compared to using a language model alone, GA LLM achieves better constraint satisfaction and higher quality solutions by combining the strengths of both components.
Large Language Models (LLMs) are increasingly assisting users in the real world, yet their reliability remains a concern. Uncertainty quantification (UQ) has been heralded as a tool to enhance human-LLM collaboration by enabling users to know when to trust LLM predictions. We argue that current practices for uncertainty quantification in LLMs are not optimal for developing useful UQ for human users making decisions in real-world tasks. Through an analysis of 40 LLM UQ methods, we identify three prevalent practices hindering the community's progress toward its goal of benefiting downstream users: 1) evaluating on benchmarks with low ecological validity; 2) considering only epistemic uncertainty; and 3) optimizing metrics that are not necessarily indicative of downstream utility. For each issue, we propose concrete user-centric practices and research directions that LLM UQ researchers should consider. Instead of hill-climbing on unrepresentative tasks using imperfect metrics, we argue that the community should adopt a more human-centered approach to LLM uncertainty quantification.
Although large language models (LLMs) are highly interactive and extendable, current approaches to ensure reliability in deployments remain mostly limited to rejecting outputs with high uncertainty in order to avoid misinformation. This conservative strategy reflects the current lack of tools to systematically distinguish and respond to different sources of uncertainty. In this paper, we advocate for the adoption of Bayesian Modeling of Experiments -- a framework that provides a coherent foundation to reason about uncertainty and clarify the reducibility of uncertainty -- for managing and proactively addressing uncertainty that arises in LLM deployments. This framework enables LLMs and their users to take contextually appropriate steps, such as requesting clarification, retrieving external information, or refining inputs. By supporting active resolution rather than passive avoidance, it opens the door to more reliable, transparent, and broadly applicable LLM systems, particularly in high-stakes, real-world settings.
Recent advances in Large Language Models (LLMs) have driven their adoption in recommender systems through Retrieval-Augmented Generation (RAG) frameworks. However, existing RAG approaches predominantly rely on flat, similarity-based retrieval that fails to leverage the rich relational structure inherent in user-item interactions. We introduce LlamaRec-LKG-RAG, a novel single-pass, end-to-end trainable framework that integrates personalized knowledge graph context into LLM-based recommendation ranking. Our approach extends the LlamaRec architecture by incorporating a lightweight user preference module that dynamically identifies salient relation paths within a heterogeneous knowledge graph constructed from user behavior and item metadata. These personalized subgraphs are seamlessly integrated into prompts for a fine-tuned Llama-2 model, enabling efficient and interpretable recommendations through a unified inference step. Comprehensive experiments on ML-100K and Amazon Beauty datasets demonstrate consistent and significant improvements over LlamaRec across key ranking metrics (MRR, NDCG, Recall). LlamaRec-LKG-RAG demonstrates the critical value of structured reasoning in LLM-based recommendations and establishes a foundation for scalable, knowledge-aware personalization in next-generation recommender systems. Code is available at~\href{https://github.com/VahidAz/LlamaRec-LKG-RAG}{repository}.
Recent studies show that in supervised fine-tuning (SFT) of large language models (LLMs), data quality matters more than quantity. While most data cleaning methods concentrate on filtering entire samples, the quality of individual tokens within a sample can vary significantly. After pre-training, even in high-quality samples, patterns or phrases that are not task-related can be redundant, uninformative, or even harmful. Continuing to fine-tune on these patterns may offer limited benefit and even degrade downstream task performance. In this paper, we investigate token quality from a noisy-label perspective and propose a generic token cleaning pipeline for SFT tasks. Our method filters out uninformative tokens while preserving those carrying key task-specific information. Specifically, we first evaluate token quality by examining the influence of model updates on each token, then apply a threshold-based separation. The token influence can be measured in a single pass with a fixed reference model or iteratively with self-evolving reference models. The benefits and limitations of both methods are analyzed theoretically by error upper bounds. Extensive experiments show that our framework consistently improves downstream performance. Code is available at https://github.com/UCSC-REAL/TokenCleaning.
Large Language Models (LLMs) exhibit strong multilingual performance despite being predominantly trained on English-centric corpora. This raises a fundamental question: How do LLMs achieve such multilingual capabilities? Focusing on languages written in non-Roman scripts, we investigate the role of Romanization - the representation of non-Roman scripts using Roman characters - as a potential bridge in multilingual processing. Using mechanistic interpretability techniques, we analyze next-token generation and find that intermediate layers frequently represent target words in Romanized form before transitioning to native script, a phenomenon we term Latent Romanization. Further, through activation patching experiments, we demonstrate that LLMs encode semantic concepts similarly across native and Romanized scripts, suggesting a shared underlying representation. Additionally, for translation into non-Roman script languages, our findings reveal that when the target language is in Romanized form, its representations emerge earlier in the model's layers compared to native script. These insights contribute to a deeper understanding of multilingual representation in LLMs and highlight the implicit role of Romanization in facilitating language transfer.
The recent emergence of multimodal large language models (LLMs) has introduced new opportunities for improving visual hazard recognition on construction sites. Unlike traditional computer vision models that rely on domain-specific training and extensive datasets, modern LLMs can interpret and describe complex visual scenes using simple natural language prompts. However, despite growing interest in their applications, there has been limited investigation into how different LLMs perform in safety-critical visual tasks within the construction domain. To address this gap, this study conducts a comparative evaluation of five state-of-the-art LLMs: Claude-3 Opus, GPT-4.5, GPT-4o, GPT-o3, and Gemini 2.0 Pro, to assess their ability to identify potential hazards from real-world construction images. Each model was tested under three prompting strategies: zero-shot, few-shot, and chain-of-thought (CoT). Zero-shot prompting involved minimal instruction, few-shot incorporated basic safety context and a hazard source mnemonic, and CoT provided step-by-step reasoning examples to scaffold model thinking. Quantitative analysis was performed using precision, recall, and F1-score metrics across all conditions. Results reveal that prompting strategy significantly influenced performance, with CoT prompting consistently producing higher accuracy across models. Additionally, LLM performance varied under different conditions, with GPT-4.5 and GPT-o3 outperforming others in most settings. The findings also demonstrate the critical role of prompt design in enhancing the accuracy and consistency of multimodal LLMs for construction safety applications. This study offers actionable insights into the integration of prompt engineering and LLMs for practical hazard recognition, contributing to the development of more reliable AI-assisted safety systems.
As large language models (LLMs) increasingly adapt and personalize to diverse sets of users, there is an increased risk of systems appropriating sociolects, i.e., language styles or dialects that are associated with specific minoritized lived experiences (e.g., African American English, Queer slang). In this work, we examine whether sociolect usage by an LLM agent affects user reliance on its outputs and user perception (satisfaction, frustration, trust, and social presence). We designed and conducted user studies where 498 African American English (AAE) speakers and 487 Queer slang speakers performed a set of question-answering tasks with LLM-based suggestions in either standard American English (SAE) or their self-identified sociolect. Our findings showed that sociolect usage by LLMs influenced both reliance and perceptions, though in some surprising ways. Results suggest that both AAE and Queer slang speakers relied more on the SAE agent, and had more positive perceptions of the SAE agent. Yet, only Queer slang speakers felt more social presence from the Queer slang agent over the SAE one, whereas only AAE speakers preferred and trusted the SAE agent over the AAE one. These findings emphasize the need to test for behavioral outcomes rather than simply assume that personalization would lead to a better and safer reliance outcome. They also highlight the nuanced dynamics of minoritized language in machine interactions, underscoring the need for LLMs to be carefully designed to respect cultural and linguistic boundaries while fostering genuine user engagement and trust.
With the rapid development of multi-cloud environments, it is increasingly important to ensure the security and reliability of intelligent monitoring systems. In this paper, we propose an anomaly detection and early warning mechanism for intelligent monitoring system in multi-cloud environment based on Large-Scale Language Model (LLM). On the basis of the existing monitoring framework, the proposed model innovatively introduces a multi-level feature extraction method, which combines the natural language processing ability of LLM with traditional machine learning methods to enhance the accuracy of anomaly detection and improve the real-time response efficiency. By introducing the contextual understanding capabilities of LLMs, the model dynamically adapts to different cloud service providers and environments, so as to more effectively detect abnormal patterns and predict potential failures. Experimental results show that the proposed model is significantly better than the traditional anomaly detection system in terms of detection accuracy and latency, and significantly improves the resilience and active management ability of cloud infrastructure.
The rapid advancements in Large Language Models (LLMs) have revolutionized various natural language processing tasks. However, the substantial size of LLMs presents significant challenges in training or fine-tuning. While parameter-efficient approaches such as low-rank adaptation (LoRA) have gained popularity, they often compromise performance compared to full-rank fine-tuning. In this paper, we propose Outlier-weighed Layerwise Sampling (OWS), a new memory-efficient fine-tuning approach, inspired by the layerwise outlier distribution of LLMs. Unlike LoRA, which adds extra adapters to all layers, OWS strategically assigns higher sampling probabilities to layers with more outliers, selectively sampling only a few layers and fine-tuning their pre-trained weights. To further increase the number of fine-tuned layers without a proportional rise in memory costs, we incorporate gradient low-rank projection, further boosting the approach's performance. Our extensive experiments across various architectures, including LLaMa2 and Mistral, demonstrate that OWS consistently outperforms baseline approaches, including full fine-tuning. Specifically, it achieves up to a 1.1% average accuracy gain on the Commonsense Reasoning benchmark, a 3.0% improvement on MMLU, and a notable 10% boost on MT-Bench, while being more memory efficient. OWS allows us to fine-tune 7B LLMs with only 21GB of memory. Our code is available at https://github.com/pixeli99/OWS.
Recent advancements in watermarking techniques have enabled the embedding of secret messages into AI-generated text (AIGT), serving as an important mechanism for AIGT detection. Existing methods typically interfere with the generation processes of large language models (LLMs) to embed signals within the generated text. However, these methods often rely on heuristic rules, which can result in suboptimal token selection and a subsequent decline in the quality of the generated content. In this paper, we introduce a plug-and-play contextual generation states-aware watermarking framework (CAW) that dynamically adjusts the embedding process. It can be seamlessly integrated with various existing watermarking methods to enhance generation quality. First, CAW incorporates a watermarking capacity evaluator, which can assess the impact of embedding messages at different token positions by analyzing the contextual generation states. Furthermore, we introduce a multi-branch pre-generation mechanism to avoid the latency caused by the proposed watermarking strategy. Building on this, CAW can dynamically adjust the watermarking process based on the evaluated watermark capacity of each token, thereby minimizing potential degradation in content quality. Extensive experiments conducted on datasets across multiple domains have verified the effectiveness of our method, demonstrating superior performance compared to various baselines in terms of both detection rate and generation quality.
Large language models (LLMs) are increasingly deployed in real-world applications, raising concerns about their security. While jailbreak attacks highlight failures under overtly harmful queries, they overlook a critical risk: incorrectly answering harmless-looking inputs can be dangerous and cause real-world harm (Implicit Harm). We systematically reformulate the LLM risk landscape through a structured quadrant perspective based on output factuality and input harmlessness, uncovering an overlooked high-risk region. To investigate this gap, we propose JailFlipBench, a benchmark aims to capture implicit harm, spanning single-modal, multimodal, and factual extension scenarios with diverse evaluation metrics. We further develop initial JailFlip attack methodologies and conduct comprehensive evaluations across multiple open-source and black-box LLMs, show that implicit harm present immediate and urgent real-world risks, calling for broader LLM safety assessments and alignment beyond conventional jailbreak paradigms.
Large language models (LLMs) demonstrate considerable proficiency in numerous coding-related tasks; however, their capabilities in detecting software vulnerabilities remain limited. This limitation primarily stems from two factors: (1) the absence of reasoning data related to vulnerabilities, which hinders the models' ability to capture underlying vulnerability patterns; and (2) their focus on learning semantic representations rather than the reason behind them, thus failing to recognize semantically similar vulnerability samples. Furthermore, the development of LLMs specialized in vulnerability detection is challenging, particularly in environments characterized by the scarcity of high-quality datasets. In this paper, we propose a novel framework ReVD that excels at mining vulnerability patterns through reasoning data synthesizing and vulnerability-specific preference optimization. Specifically, we construct forward and backward reasoning processes for vulnerability and corresponding fixed code, ensuring the synthesis of high-quality reasoning data. Moreover, we design the triplet supervised fine-tuning followed by curriculum online preference optimization for enabling ReVD to better understand vulnerability patterns. The extensive experiments conducted on PrimeVul and SVEN datasets demonstrate that ReVD sets new state-of-the-art for LLM-based software vulnerability detection, e.g., 12.24\%-22.77\% improvement in the accuracy. The source code and data are available at https://github.com/Xin-Cheng-Wen/PO4Vul.
Large Language Models (LLMs) show strong collaborative performance in multi-agent systems with predefined roles and workflows. However, in open-ended environments lacking coordination rules, agents tend to act in self-interested ways. The central challenge in achieving coordination lies in credit assignment -- fairly evaluating each agent's contribution and designing pricing mechanisms that align their heterogeneous goals. This problem is critical as LLMs increasingly participate in complex human-AI collaborations, where fair compensation and accountability rely on effective pricing mechanisms. Inspired by how human societies address similar coordination challenges (e.g., through temporary collaborations such as employment or subcontracting), we propose a cooperative workflow, Shapley-Coop. Shapley-Coop integrates Shapley Chain-of-Thought -- leveraging marginal contributions as a principled basis for pricing -- with structured negotiation protocols for effective price matching, enabling LLM agents to coordinate through rational task-time pricing and post-task reward redistribution. This approach aligns agent incentives, fosters cooperation, and maintains autonomy. We evaluate Shapley-Coop across two multi-agent games and a software engineering simulation, demonstrating that it consistently enhances LLM agent collaboration and facilitates equitable credit assignment. These results highlight the effectiveness of Shapley-Coop's pricing mechanisms in accurately reflecting individual contributions during task execution.
Video large language models have not yet been widely deployed, largely due to their tendency to hallucinate. Typical benchmarks for Video-LLMs rely simply on multiple-choice questions. Unfortunately, VideoLLMs hallucinate far more aggressively on freeform text generation tasks like video captioning than they do on multiple choice verification tasks. To address this weakness, we propose ARGUS, a VideoLLM benchmark that measures freeform video captioning performance. By comparing VideoLLM outputs to human ground truth captions, ARGUS quantifies dual metrics. First, we measure the rate of hallucinations in the form of incorrect statements about video content or temporal relationships. Second, we measure the rate at which the model omits important descriptive details. Together, these dual metrics form a comprehensive view of video captioning performance.
The remarkable success of Large Language Models (LLMs) has illuminated a promising pathway toward achieving Artificial General Intelligence for both academic and industrial communities, owing to their unprecedented performance across various applications. As LLMs continue to gain prominence in both research and commercial domains, their security and safety implications have become a growing concern, not only for researchers and corporations but also for every nation. Currently, existing surveys on LLM safety primarily focus on specific stages of the LLM lifecycle, e.g., deployment phase or fine-tuning phase, lacking a comprehensive understanding of the entire "lifechain" of LLMs. To address this gap, this paper introduces, for the first time, the concept of "full-stack" safety to systematically consider safety issues throughout the entire process of LLM training, deployment, and eventual commercialization. Compared to the off-the-shelf LLM safety surveys, our work demonstrates several distinctive advantages: (I) Comprehensive Perspective. We define the complete LLM lifecycle as encompassing data preparation, pre-training, post-training, deployment and final commercialization. To our knowledge, this represents the first safety survey to encompass the entire lifecycle of LLMs. (II) Extensive Literature Support. Our research is grounded in an exhaustive review of over 800+ papers, ensuring comprehensive coverage and systematic organization of security issues within a more holistic understanding. (III) Unique Insights. Through systematic literature analysis, we have developed reliable roadmaps and perspectives for each chapter. Our work identifies promising research directions, including safety in data generation, alignment techniques, model editing, and LLM-based agent systems. These insights provide valuable guidance for researchers pursuing future work in this field.
Alpha mining, a critical component in quantitative investment, focuses on discovering predictive signals for future asset returns in increasingly complex financial markets. However, the pervasive issue of alpha decay, where factors lose their predictive power over time, poses a significant challenge for alpha mining. Traditional methods like genetic programming face rapid alpha decay from overfitting and complexity, while approaches driven by Large Language Models (LLMs), despite their promise, often rely too heavily on existing knowledge, creating homogeneous factors that worsen crowding and accelerate decay. To address this challenge, we propose AlphaAgent, an autonomous framework that effectively integrates LLM agents with ad hoc regularizations for mining decay-resistant alpha factors. AlphaAgent employs three key mechanisms: (i) originality enforcement through a similarity measure based on abstract syntax trees (ASTs) against existing alphas, (ii) hypothesis-factor alignment via LLM-evaluated semantic consistency between market hypotheses and generated factors, and (iii) complexity control via AST-based structural constraints, preventing over-engineered constructions that are prone to overfitting. These mechanisms collectively guide the alpha generation process to balance originality, financial rationale, and adaptability to evolving market conditions, mitigating the risk of alpha decay. Extensive evaluations show that AlphaAgent outperforms traditional and LLM-based methods in mitigating alpha decay across bull and bear markets, consistently delivering significant alpha in Chinese CSI 500 and US S&P 500 markets over the past four years. Notably, AlphaAgent showcases remarkable resistance to alpha decay, elevating the potential for yielding powerful factors.
Large Language Models (LLMs) excel in traditional natural language processing tasks but struggle with problems that require complex domain-specific calculations or simulations. While equipping LLMs with external tools to build LLM-based agents can enhance their capabilities, existing approaches lack the flexibility to address diverse and ever-evolving user queries in open domains. Currently, there is also no existing dataset that evaluates LLMs on open-domain knowledge that requires tools to solve. To this end, we introduce OpenAct benchmark to evaluate the open-domain task-solving capability, which is built on human expert consultation and repositories in GitHub. It comprises 339 questions spanning 7 diverse domains that need to be solved with domain-specific methods. In our experiments, even state-of-the-art LLMs and LLM-based agents demonstrate unsatisfactory success rates, underscoring the need for a novel approach. Furthermore, we present OpenAgent, a novel LLM-based agent system that can tackle evolving queries in open domains through autonomously integrating specialized tools from GitHub. OpenAgent employs 1) a hierarchical framework where specialized agents handle specific tasks and can assign tasks to inferior agents, 2) a bi-level experience learning mechanism to learn from both humans' and its own experiences to tackle tool flaws. Experiments demonstrate its superior effectiveness and efficiency, which significantly outperforms baselines. Our data and code are open-source at https://github.com/OpenBMB/OpenAct.
This study introduces "CosmoAgent," an innovative artificial intelligence system that utilizes Large Language Models (LLMs) to simulate complex interactions between human and extraterrestrial civilizations. This paper introduces a mathematical model for quantifying the levels of civilization development and further employs a state transition matrix approach to evaluate their trajectories. Through this methodology, our study quantitatively analyzes the growth trajectories of civilizations, providing insights into future decision-making at critical points of growth and saturation. Furthermore, this paper acknowledges the vast diversity of potential living conditions across the universe, which could foster unique cosmologies, ethical codes, and worldviews among different civilizations. Recognizing the Earth-centric bias inherent in current LLM designs, we propose the novel concept of using LLM agents with diverse ethical paradigms and simulating interactions between entities with distinct moral principles. This innovative research not only introduces a novel method for comprehending potential inter-civilizational dynamics but also holds practical value in enabling entities with divergent value systems to strategize, prevent conflicts, and engage in games under conditions of asymmetric information. The accompanying code is available at https://github.com/MingyuJ666/Simulating-Alien-Civilizations-with-LLM-based-Agents.
Role-playing has emerged as an effective technique for enhancing the reasoning capabilities of large language models (LLMs). However, existing methods primarily rely on prompt engineering, which often lacks stability and interpretability. In this paper, we introduce Sparse Autoencoder Role-Playing Steering (SRPS), a novel framework that identifies and manipulates internal model features associated with role-playing behavior. Our approach extracts latent representations from role-play prompts, selects the most relevant features based on activation patterns, and constructs a steering vector that can be injected into the model's residual stream with controllable intensity. Our method enables fine-grained control over role-specific behavior and offers insights into how role information influences internal model activations. Extensive experiments across various reasoning benchmarks and model sizes demonstrate consistent performance gains. Notably, in the zero-shot chain-of-thought (CoT) setting, the accuracy of Llama3.1-8B on CSQA improves from 31.86% to 39.80%, while Gemma2-9B on SVAMP increases from 37.50% to 45.10%. These results highlight the potential of SRPS to enhance reasoning ability in LLMs, providing better interpretability and stability compared to traditional prompt-based role-playing.
We present JavelinGuard, a suite of low-cost, high-performance model architectures designed for detecting malicious intent in Large Language Model (LLM) interactions, optimized specifically for production deployment. Recent advances in transformer architectures, including compact BERT(Devlin et al. 2019) variants (e.g., ModernBERT (Warner et al. 2024)), allow us to build highly accurate classifiers with as few as approximately 400M parameters that achieve rapid inference speeds even on standard CPU hardware. We systematically explore five progressively sophisticated transformer-based architectures: Sharanga (baseline transformer classifier), Mahendra (enhanced attention-weighted pooling with deeper heads), Vaishnava and Ashwina (hybrid neural ensemble architectures), and Raudra (an advanced multi-task framework with specialized loss functions). Our models are rigorously benchmarked across nine diverse adversarial datasets, including popular sets like the NotInject series, BIPIA, Garak, ImprovedLLM, ToxicChat, WildGuard, and our newly introduced JavelinBench, specifically crafted to test generalization on challenging borderline and hard-negative cases. Additionally, we compare our architectures against leading open-source guardrail models as well as large decoder-only LLMs such as gpt-4o, demonstrating superior cost-performance trade-offs in terms of accuracy, and latency. Our findings reveal that while Raudra's multi-task design offers the most robust performance overall, each architecture presents unique trade-offs in speed, interpretability, and resource requirements, guiding practitioners in selecting the optimal balance of complexity and efficiency for real-world LLM security applications.
Multi-agent frameworks can substantially boost the reasoning power of large language models (LLMs), but they typically incur heavy computational costs and lack convergence guarantees. To overcome these challenges, we recast multi-LLM coordination as an incomplete-information game and seek a Bayesian Nash equilibrium (BNE), in which each agent optimally responds to its probabilistic beliefs about the strategies of others. We introduce Efficient Coordination via Nash Equilibrium (ECON), a hierarchical reinforcement-learning paradigm that marries distributed reasoning with centralized final output. Under ECON, each LLM independently selects responses that maximize its expected reward, conditioned on its beliefs about co-agents, without requiring costly inter-agent exchanges. We mathematically prove that ECON attains a markedly tighter regret bound than non-equilibrium multi-agent schemes. Empirically, ECON outperforms existing multi-LLM approaches by 11.2% on average across six benchmarks spanning complex reasoning and planning tasks. Further experiments demonstrate ECON's ability to flexibly incorporate additional models, confirming its scalability and paving the way toward larger, more powerful multi-LLM ensembles. The code is publicly available at: https://github.com/tmlr-group/ECON.
Multimodal retrieval augmented generation (RAG) plays a crucial role in domains such as knowledge-based visual question answering (KB-VQA), where external knowledge is needed to answer a question. However, existing multimodal LLMs (MLLMs) are not designed for context-augmented generation, limiting their effectiveness in such tasks. While synthetic data generation has recently gained attention for training MLLMs, its application for context-augmented generation remains underexplored. To address this gap, we introduce SK-VQA, a large-scale synthetic multimodal dataset containing over 2 million visual question-answer pairs, each associated with context documents containing information necessary to determine the final answer. Compared to previous datasets, SK-VQA contains 11x more unique questions, exhibits greater domain diversity, and covers a broader spectrum of image sources. Through human evaluations, we confirm the high quality of the generated question-answer pairs and their contextual relevance. Extensive experiments show that SK-VQA serves both as a challenging KB-VQA benchmark and as an effective training resource for adapting MLLMs to context-augmented generation. Our results further indicate that models trained on SK-VQA demonstrate enhanced generalization in both context-aware VQA and multimodal RAG settings. SK-VQA is publicly available via Hugging Face Hub.
Large language models (LLMs) are increasingly being used for complex research tasks such as literature review, idea generation, and scientific paper analysis, yet their ability to truly understand and process the intricate relationships within complex research papers, such as the logical links between claims and supporting evidence remains largely unexplored. In this study, we present CLAIM-BENCH, a comprehensive benchmark for evaluating LLMs' capabilities in scientific claim-evidence extraction and validation, a task that reflects deeper comprehension of scientific argumentation. We systematically compare three approaches which are inspired by divide and conquer approaches, across six diverse LLMs, highlighting model-specific strengths and weaknesses in scientific comprehension. Through evaluation involving over 300 claim-evidence pairs across multiple research domains, we reveal significant limitations in LLMs' ability to process complex scientific content. Our results demonstrate that closed-source models like GPT-4 and Claude consistently outperform open-source counterparts in precision and recall across claim-evidence identification tasks. Furthermore, strategically designed three-pass and one-by-one prompting approaches significantly improve LLMs' abilities to accurately link dispersed evidence with claims, although this comes at increased computational cost. CLAIM-BENCH sets a new standard for evaluating scientific comprehension in LLMs, offering both a diagnostic tool and a path forward for building systems capable of deeper, more reliable reasoning across full-length papers.
Large language models (LLMs) are increasingly used to extract clinical data from electronic health records (EHRs), offering significant improvements in scalability and efficiency for real-world data (RWD) curation in oncology. However, the adoption of LLMs introduces new challenges in ensuring the reliability, accuracy, and fairness of extracted data, which are essential for research, regulatory, and clinical applications. Existing quality assurance frameworks for RWD and artificial intelligence do not fully address the unique error modes and complexities associated with LLM-extracted data. In this paper, we propose a comprehensive framework for evaluating the quality of clinical data extracted by LLMs. The framework integrates variable-level performance benchmarking against expert human abstraction, automated verification checks for internal consistency and plausibility, and replication analyses comparing LLM-extracted data to human-abstracted datasets or external standards. This multidimensional approach enables the identification of variables most in need of improvement, systematic detection of latent errors, and confirmation of dataset fitness-for-purpose in real-world research. Additionally, the framework supports bias assessment by stratifying metrics across demographic subgroups. By providing a rigorous and transparent method for assessing LLM-extracted RWD, this framework advances industry standards and supports the trustworthy use of AI-powered evidence generation in oncology research and practice.
Large language models(LLMs) like Gemini are becoming common tools for supporting student writing. But most of their feedback is based only on the final essay missing important context about how that text was written. In this paper, we explore whether using writing process data, collected through keystroke logging and periodic snapshots, can help LLMs give feedback that better reflects how learners think and revise while writing. We built a digital writing tool that captures both what students type and how their essays evolve over time. Twenty students used this tool to write timed essays, which were then evaluated in two ways: (i) LLM generated feedback using both the final essay and the full writing trace, and (ii) After the task, students completed surveys about how useful and relatable they found the feedback. Early results show that learners preferred the process-aware LLM feedback, finding it more in tune with their own thinking. We also found that certain types of edits, like adding new content or reorganizing paragraphs, aligned closely with higher scores in areas like coherence and elaboration. Our findings suggest that making LLMs more aware of the writing process can lead to feedback that feels more meaningful, personal, and supportive.
Both text-to-image generation and large language models (LLMs) have made significant advancements. However, many text-to-image models still employ the somewhat outdated T5 and CLIP as their text encoders. In this work, we investigate the effectiveness of using modern decoder-only LLMs as text encoders for text-to-image diffusion models. We build a standardized training and evaluation pipeline that allows us to isolate and evaluate the effect of different text embeddings. We train a total of 27 text-to-image models with 12 different text encoders to analyze the critical aspects of LLMs that could impact text-to-image generation, including the approaches to extract embeddings, different LLMs variants, and model sizes. Our experiments reveal that the de facto way of using last-layer embeddings as conditioning leads to inferior performance. Instead, we explore embeddings from various layers and find that using layer-normalized averaging across all layers significantly improves alignment with complex prompts. Most LLMs with this conditioning outperform the baseline T5 model, showing enhanced performance in advanced visio-linguistic reasoning skills.
Large Language Models (LLMs) pose significant privacy risks, potentially leaking training data due to implicit memorization. Existing privacy attacks primarily focus on membership inference attacks (MIAs) or data extraction attacks, but reconstructing specific personally identifiable information (PII) in LLMs' training data remains challenging. In this paper, we propose R.R. (Recollect and Rank), a novel two-step privacy stealing attack that enables attackers to reconstruct PII entities from scrubbed training data where the PII entities have been masked. In the first stage, we introduce a prompt paradigm named recollection, which instructs the LLM to repeat a masked text but fill in masks. Then we can use PII identifiers to extract recollected PII candidates. In the second stage, we design a new criterion to score each PII candidate and rank them. Motivated by membership inference, we leverage the reference model as a calibration to our criterion. Experiments across three popular PII datasets demonstrate that the R.R. achieves better PII identification performance than baselines. These results highlight the vulnerability of LLMs to PII leakage even when training data has been scrubbed. We release our code and datasets at GitHub.
Large Language Models (LLMs) have revolutionized artificial intelligence with capabilities in reasoning, coding, and communication, driving innovation across industries. Their true potential depends on effective alignment to ensure correct, trustworthy and ethical behavior, addressing challenges like misinformation, hallucinations, bias and misuse. While existing Reinforcement Learning (RL)-based alignment methods are notoriously complex, direct optimization approaches offer a simpler alternative. In this work, we introduce a novel direct optimization approach for LLM alignment by drawing on established Information Retrieval (IR) principles. We present a systematic framework that bridges LLM alignment and IR methodologies, mapping LLM generation and reward models to IR's retriever-reranker paradigm. Building on this foundation, we propose LLM Alignment as Retriever Preference Optimization (LarPO), a new alignment method that enhances overall alignment quality. Extensive experiments validate LarPO's effectiveness with 38.9 % and 13.7 % averaged improvement on AlpacaEval2 and MixEval-Hard respectively. Our work opens new avenues for advancing LLM alignment by integrating IR foundations, offering a promising direction for future research.
Enabling large language models (LLMs) to unlearn knowledge and capabilities acquired during training has proven vital for ensuring compliance with data regulations and promoting ethical practices in generative AI. Although there are growing interests in developing various unlearning algorithms, it remains unclear how to best formulate the unlearning problem. The most popular formulation uses a weighted sum of forget and retain loss, but it often leads to performance degradation due to the inherent trade-off between forget and retain losses. In this work, we argue that it is important to model the hierarchical structure of the unlearning problem, where the forget problem (which \textit{unlearns} certain knowledge and/or capabilities) takes priority over the retain problem (which preserves model utility). This hierarchical structure naturally leads to a bi-level optimization formulation where the lower-level objective focuses on minimizing the forget loss, while the upper-level objective aims to maintain the model's utility. Based on this new formulation, we propose a novel algorithm, termed Bi-Level UnleaRning (\texttt{BLUR}), which not only possesses strong theoretical guarantees but more importantly, delivers superior performance. In particular, our extensive experiments demonstrate that \texttt{BLUR} consistently outperforms all the state-of-the-art algorithms across various unlearning tasks, models, and metrics. Codes are available at https://github.com/OptimAI-Lab/BLURLLMUnlearning.
Large language models have demonstrated impressive reasoning capabilities, yet they often suffer from inefficiencies due to unnecessarily verbose or redundant outputs. While many works have explored reinforcement learning (RL) to enhance reasoning abilities, most primarily focus on improving accuracy, with limited attention to reasoning efficiency. Some existing approaches introduce direct length-based rewards to encourage brevity, but this often leads to noticeable drops in accuracy. In this paper, we propose Bingo, an RL framework that advances length-based reward design to boost efficient reasoning. Bingo incorporates two key mechanisms: a significance-aware length reward, which gradually guides the model to reduce only insignificant tokens, and a dynamic length reward, which initially encourages elaborate reasoning for hard questions but decays over time to improve overall efficiency. Experiments across multiple reasoning benchmarks show that Bingo improves both accuracy and efficiency. It outperforms the vanilla reward and several other length-based reward baselines in RL, achieving a favorable trade-off between accuracy and efficiency. These results underscore the potential of training LLMs explicitly for efficient reasoning.
Large Language Models (LLMs) demonstrate impressive general-purpose reasoning and problem-solving abilities. However, they struggle with executing complex, long-horizon workflows that demand strict adherence to Standard Operating Procedures (SOPs), a critical requirement for real-world industrial automation. Despite this need, there is a lack of public benchmarks that reflect the complexity, structure, and domain-specific nuances of SOPs. To address this, we present three main contributions. First, we introduce a synthetic data generation framework to create realistic, industry-grade SOPs that rigorously test the planning, reasoning, and tool-use capabilities of LLM-based agents. Second, using this framework, we develop SOP-Bench, a benchmark of over 1,800 tasks across 10 industrial domains, each with APIs, tool interfaces, and human-validated test cases. Third, we evaluate two prominent agent architectures: Function-Calling and ReAct Agents, on SOP-Bench, observing average success rates of only 27% and 48%, respectively. Remarkably, when the tool registry is much larger than necessary, agents invoke incorrect tools nearly 100% of the time. These findings underscore a substantial gap between current agentic capabilities of LLMs and the demands of automating real-world SOPs. Performance varies significantly by task and domain, highlighting the need for domain-specific benchmarking and architectural choices before deployment. SOP-Bench is publicly available at http://sop-bench.s3-website-us-west-2.amazonaws.com/. We also release the prompts underpinning the data generation framework to support new domain-specific SOP benchmarks. We invite the community to extend SOP-Bench with SOPs from their industrial domains.
In recent years, Large Language Models (LLMs) rapidly gained popularity across all parts of society, including education. After initial skepticism and bans, many schools have chosen to embrace this new technology by integrating it into their curricula in the form of virtual tutors and teaching assistants. However, neither the companies developing this technology nor the public institutions involved in its implementation have set up a formal system to collect feedback from the stakeholders impacted by them. In this paper, we argue that understanding the perceptions of those directly or indirectly impacted by LLMs in the classroom, including parents and school staff, is essential for ensuring responsible use of AI in this critical domain. Our contributions are two-fold. First, we propose the Contextualized Perceptions for the Adoption of LLMs in Education (Co-PALE) framework, which can be used to systematically elicit perceptions and inform whether and how LLM-based tools should be designed, developed, and deployed in the classroom. Second, we explain how our framework can be used to ground specific rubrics for eliciting perceptions of the relevant stakeholders in view of specific goals and context of implementation. Overall, Co-PALE is a practical step toward helping educational agents, policymakers, researchers, and technologists ensure the responsible and effective deployment of LLM-based systems across diverse learning contexts.
Post-processing is crucial for the automatic evaluation of LLMs in fill-in-the-middle (FIM) code generation due to the frequent presence of extraneous code in raw outputs. This extraneous generation suggests a lack of awareness regarding output boundaries, requiring truncation for effective evaluation. The determination of an optimal truncation strategy, however, often proves intricate, particularly when the scope includes several programming languages. This study investigates the necessity of post-processing instruction-tuned LLM outputs. Our findings reveal that supervised fine-tuning significantly enhances FIM code generation, enabling LLMs to generate code that seamlessly integrates with the surrounding context. Evaluating our fine-tuned \texttt{Qwen2.5-Coder} (base and instruct) models on HumanEval Infilling and SAFIM benchmarks demonstrates improved performances without post-processing, especially when the \emph{middle} consist of complete lines. However, post-processing of the LLM outputs remains necessary when the \emph{middle} is a random span of code.
While large language models (LLMs) are driving the rapid advancement of artificial intelligence, effectively and reliably training these large models remains one of the field's most significant challenges. To address this challenge, we propose POET, a novel reParameterized training algorithm that uses Orthogonal Equivalence Transformation to optimize neurons. Specifically, POET reparameterizes each neuron with two learnable orthogonal matrices and a fixed random weight matrix. Because of its provable preservation of spectral properties of weight matrices, POET can stably optimize the objective function with improved generalization. We further develop efficient approximations that make POET flexible and scalable for training large-scale neural networks. Extensive experiments validate the effectiveness and scalability of POET in training LLMs.
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 LLMs are now increasingly deployed without careful review in assistant and agentic roles. Code for our experiments is available at https://github.com/pradyuprasad/llms_overconfidence
While Large Language Models (LLMs) have demonstrated significant advancements in reasoning and agent-based problem-solving, current evaluation methodologies fail to adequately assess their capabilities: existing benchmarks either rely on closed-ended questions prone to saturation and memorization, or subjective comparisons that lack consistency and rigor. In this work, we introduce HeuriGym, an agentic framework designed for evaluating heuristic algorithms generated by LLMs for combinatorial optimization problems, characterized by clearly defined objectives and expansive solution spaces. HeuriGym empowers LLMs to propose heuristics, receive evaluative feedback via code execution, and iteratively refine their solutions. We evaluate nine state-of-the-art models on nine problems across domains such as computer systems, logistics, and biology, exposing persistent limitations in tool use, planning, and adaptive reasoning. To quantify performance, we propose the Quality-Yield Index (QYI), a metric that captures both solution pass rate and quality. Even top models like GPT-o4-mini-high and Gemini-2.5-Pro attain QYI scores of only 0.6, well below the expert baseline of 1. Our open-source benchmark aims to guide the development of LLMs toward more effective and realistic problem-solving in scientific and engineering domains.
As the Large-Language-Model-driven (LLM-driven) Artificial Intelligence (AI) bots became popular, people realized their strong potential in Task-Oriented Dialogue (TOD). However, bots relying wholly on LLMs are unreliable in their knowledge, and whether they can finally produce a correct outcome for the task is not guaranteed. The collaboration among these agents also remains a challenge, since the necessary information to convey is unclear, and the information transfer is by prompts: unreliable, and malicious knowledge is easy to inject. With the help of knowledge representation and reasoning tools such as Answer Set Programming (ASP), conversational agents can be built safely and reliably, and communication among the agents made more reliable as well. We propose a Manager-Customer-Service Dual-Agent paradigm, where ASP-driven bots share the same knowledge base and complete their assigned tasks independently. The agents communicate with each other through the knowledge base, ensuring consistency. The knowledge and information conveyed are encapsulated and invisible to the users, ensuring the security of information transmission. To illustrate the dual-agent conversational paradigm, we have constructed AutoManager, a collaboration system for managing the drive-through window of a fast-food restaurant such as Taco Bell in the US. In AutoManager, the customer service bot takes the customer's order while the manager bot manages the menu and food supply. We evaluated our AutoManager system and compared it with the real-world Taco Bell Drive-Thru AI Order Taker, and the results show that our method is more reliable.
Reinforcement learning (RL) has recently demonstrated strong potential in enhancing the reasoning capabilities of large language models (LLMs). Particularly, the "Zero" reinforcement learning introduced by Deepseek-R1-Zero, enables direct RL training of base LLMs without relying on an intermediate supervised fine-tuning stage. Despite these advancements, current works for LLM reasoning mainly focus on mathematical and coding domains, largely due to data abundance and the ease of answer verification. This limits the applicability and generalization of such models to broader domains, where questions often have diverse answer representations, and data is more scarce. In this paper, we propose General-Reasoner, a novel training paradigm designed to enhance LLM reasoning capabilities across diverse domains. Our key contributions include: (1) constructing a large-scale, high-quality dataset of questions with verifiable answers curated by web crawling, covering a wide range of disciplines; and (2) developing a generative model-based answer verifier, which replaces traditional rule-based verification with the capability of chain-of-thought and context-awareness. We train a series of models and evaluate them on a wide range of datasets covering wide domains like physics, chemistry, finance, electronics etc. Our comprehensive evaluation across these 12 benchmarks (e.g. MMLU-Pro, GPQA, SuperGPQA, TheoremQA, BBEH and MATH AMC) demonstrates that General-Reasoner outperforms existing baseline methods, achieving robust and generalizable reasoning performance while maintaining superior effectiveness in mathematical reasoning tasks.
While generalization over tasks from easy to hard is crucial to profile language models (LLMs), the datasets with fine-grained difficulty annotations for each problem across a broad range of complexity are still blank. Aiming to address this limitation, we present Easy2Hard-Bench, a consistently formatted collection of 6 benchmark datasets spanning various domains, such as mathematics and programming problems, chess puzzles, and reasoning questions. Each problem within these datasets is annotated with numerical difficulty scores. To systematically estimate problem difficulties, we collect abundant performance data on attempts to each problem by humans in the real world or LLMs on the prominent leaderboard. Leveraging the rich performance data, we apply well-established difficulty ranking systems, such as Item Response Theory (IRT) and Glicko-2 models, to uniformly assign numerical difficulty scores to problems. Moreover, datasets in Easy2Hard-Bench distinguish themselves from previous collections by a higher proportion of challenging problems. Through extensive experiments with six state-of-the-art LLMs, we provide a comprehensive analysis of their performance and generalization capabilities across varying levels of difficulty, with the aim of inspiring future research in LLM generalization. The datasets are available at https://huggingface.co/datasets/furonghuang-lab/Easy2Hard-Bench.
This paper introduces MiniCPM4, a highly efficient large language model (LLM) designed explicitly for end-side devices. We achieve this efficiency through systematic innovation in four key dimensions: model architecture, training data, training algorithms, and inference systems. Specifically, in terms of model architecture, we propose InfLLM v2, a trainable sparse attention mechanism that accelerates both prefilling and decoding phases for long-context processing. Regarding training data, we propose UltraClean, an efficient and accurate pre-training data filtering and generation strategy, and UltraChat v2, a comprehensive supervised fine-tuning dataset. These datasets enable satisfactory model performance to be achieved using just 8 trillion training tokens. Regarding training algorithms, we propose ModelTunnel v2 for efficient pre-training strategy search, and improve existing post-training methods by introducing chunk-wise rollout for load-balanced reinforcement learning and data-efficient tenary LLM, BitCPM. Regarding inference systems, we propose CPM.cu that integrates sparse attention, model quantization, and speculative sampling to achieve efficient prefilling and decoding. To meet diverse on-device requirements, MiniCPM4 is available in two versions, with 0.5B and 8B parameters, respectively. Sufficient evaluation results show that MiniCPM4 outperforms open-source models of similar size across multiple benchmarks, highlighting both its efficiency and effectiveness. Notably, MiniCPM4-8B demonstrates significant speed improvements over Qwen3-8B when processing long sequences. Through further adaptation, MiniCPM4 successfully powers diverse applications, including trustworthy survey generation and tool use with model context protocol, clearly showcasing its broad usability.
Large language models (LLMs) increasingly demonstrate signs of conceptual understanding, yet much of their internal knowledge remains latent, loosely structured, and difficult to access or evaluate. We propose self-questioning as a lightweight and scalable strategy to improve LLMs' understanding, particularly in domains where success depends on fine-grained semantic distinctions. To evaluate this approach, we introduce a challenging new benchmark of 1.3 million post-2015 computer science patent pairs, characterized by dense technical jargon and strategically complex writing. The benchmark centers on a pairwise differentiation task: can a model distinguish between closely related but substantively different inventions? We show that compared to placebo scientific information, prompting LLMs to generate and answer their own questions - targeting the background knowledge required for the task - significantly improves performance. These self-generated questions and answers activate otherwise underutilized internal knowledge. Allowing LLMs to retrieve answers from external scientific texts further enhances performance, suggesting that model knowledge is compressed and lacks the full richness of the training data. We also find that chain-of-thought prompting and self-questioning converge, though self-questioning remains more effective for improving understanding of technical concepts. Notably, we uncover an asymmetry in prompting: smaller models often generate more fundamental, more open-ended, better-aligned questions for mid-sized models than large models do, revealing a new strategy for cross-model collaboration. Altogether, our findings establish self-questioning as both a practical mechanism for automatically improving LLM comprehension, especially in domains with sparse and underrepresented knowledge, and a diagnostic probe of how internal and external knowledge are organized.
In-Context Reinforcement Learning (ICRL) is a frontier paradigm to solve Reinforcement Learning (RL) problems in the foundation model era. While ICRL capabilities have been demonstrated in transformers through task-specific training, the potential of Large Language Models (LLMs) out-of-the-box remains largely unexplored. This paper investigates whether LLMs can generalize cross-domain to perform ICRL under the problem of Dueling Bandits (DB), a stateless preference-based RL setting. We find that the top-performing LLMs exhibit a notable zero-shot capacity for relative decision-making, which translates to low short-term weak regret across all DB environment instances by quickly including the best arm in duels. However, an optimality gap still exists between LLMs and classic DB algorithms in terms of strong regret. LLMs struggle to converge and consistently exploit even when explicitly prompted to do so, and are sensitive to prompt variations. To bridge this gap, we propose an agentic flow framework: LLM with Enhanced Algorithmic Dueling (LEAD), which integrates off-the-shelf DB algorithm support with LLM agents through fine-grained adaptive interplay. We show that LEAD has theoretical guarantees inherited from classic DB algorithms on both weak and strong regret. We validate its efficacy and robustness even with noisy and adversarial prompts. The design of such an agentic framework sheds light on how to enhance the trustworthiness of general-purpose LLMs generalized to in-context decision-making tasks.
LLMs are bound to transform healthcare with advanced decision support and flexible chat assistants. However, LLMs are prone to generate inaccurate medical content. To ground LLMs in high-quality medical knowledge, LLMs have been equipped with external knowledge via RAG, where unstructured medical knowledge is split into small text chunks that can be selectively retrieved and integrated into the LLMs context. Yet, existing RAG pipelines rely on raw, unstructured medical text, which can be noisy, uncurated and difficult for LLMs to effectively leverage. Systematic approaches to organize medical knowledge to best surface it to LLMs are generally lacking. To address these challenges, we introduce MIRIAD, a large-scale, curated corpus of 5,821,948 medical QA pairs, each rephrased from and grounded in a passage from peer-reviewed medical literature using a semi-automated pipeline combining LLM generation, filtering, grounding, and human annotation. Unlike prior medical corpora, which rely on unstructured text, MIRIAD encapsulates web-scale medical knowledge in an operationalized query-response format, which enables more targeted retrieval. Experiments on challenging medical QA benchmarks show that augmenting LLMs with MIRIAD improves accuracy up to 6.7% compared to unstructured RAG baselines with the same source corpus and with the same amount of retrieved text. Moreover, MIRIAD improved the ability of LLMs to detect medical hallucinations by 22.5 to 37% (increase in F1 score). We further introduce MIRIAD-Atlas, an interactive map of MIRIAD spanning 56 medical disciplines, enabling clinical users to visually explore, search, and refine medical knowledge. MIRIAD promises to unlock a wealth of down-stream applications, including medical information retrievers, enhanced RAG applications, and knowledge-grounded chat interfaces, which ultimately enables more reliable LLM applications in healthcare.
Large Language Model (LLM) unlearning aims to erase or suppress undesirable knowledge within the model, offering promise for controlling harmful or private information to prevent misuse. However, recent studies highlight its limited efficacy in real-world scenarios, hindering practical adoption. In this study, we identify a pervasive issue underlying many downstream failures: the effectiveness of existing unlearning methods heavily depends on the form of training samples and frequently fails to generalize to alternate expressions of the same knowledge. We formally characterize this problem as Form-Dependent Bias and systematically investigate its specific manifestation patterns across various downstream tasks. To quantify its prevalence and support future research, we introduce ORT, a novel benchmark designed to evaluate the robustness of unlearning methods against variations in knowledge expression. Results reveal that Form-Dependent Bias is both widespread and severe among current techniques. We argue that LLM unlearning should be form-independent to address the endless forms of downstream tasks encountered in real-world security-critical scenarios. Towards this goal, we introduce Rank-one Concept Redirection (ROCR), a novel training-free method, as a promising solution path. ROCR performs unlearning by targeting the invariants in downstream tasks, specifically the activated dangerous concepts. It is capable of modifying model parameters within seconds to redirect the model's perception of a specific unlearning target concept to another harmless concept. Extensive experiments demonstrate that ROCR significantly improves unlearning effectiveness compared to traditional methods while generating highly natural outputs.
This paper provides an in-depth technical analysis and implementation methodology of the open-source Agent-to-Agent (A2A) protocol developed by Google and the Model Context Protocol (MCP) introduced by Anthropic. While the evolution of LLM-based autonomous agents is rapidly accelerating, efficient interactions among these agents and their integration with external systems remain significant challenges. In modern AI systems, collaboration between autonomous agents and integration with external tools have become essential elements for building practical AI applications. A2A offers a standardized communication method that enables agents developed in heterogeneous environments to collaborate effectively, while MCP provides a structured I/O framework for agents to connect with external tools and resources. Prior studies have focused primarily on the features and applications of either A2A or MCP individually. In contrast, this study takes an integrated approach, exploring how the two protocols can complement each other to address interoperability issues and facilitate efficient collaboration within complex agent ecosystems.
Recent advancements in Reinforcement Learning with Human Feedback (RLHF) have significantly impacted the alignment of Large Language Models (LLMs). The sensitivity of reinforcement learning algorithms such as Proximal Policy Optimization (PPO) has led to new line work on Direct Policy Optimization (DPO), which treats RLHF in a supervised learning framework. The increased practical use of these RLHF methods warrants an analysis of their vulnerabilities. In this work, we investigate the vulnerabilities of DPO to poisoning attacks under different scenarios and compare the effectiveness of preference poisoning, a first of its kind. We comprehensively analyze DPO's vulnerabilities under different types of attacks, i.e., backdoor and non-backdoor attacks, and different poisoning methods across a wide array of language models, i.e., LLama 7B, Mistral 7B, and Gemma 7B. We find that unlike PPO-based methods, which, when it comes to backdoor attacks, require at least 4\% of the data to be poisoned to elicit harmful behavior, we exploit the true vulnerabilities of DPO more simply so we can poison the model with only as much as 0.5\% of the data. We further investigate the potential reasons behind the vulnerability and how well this vulnerability translates into backdoor vs non-backdoor attacks.
Recent studies have shown that large language models (LLMs), especially smaller ones, often lack robustness in their reasoning. I.e., they tend to experience performance drops when faced with distribution shifts, such as changes to numerical or nominal variables, or insertions of distracting clauses. A possible strategy to address this involves generating synthetic data to further "instantiate" reasoning problems on potential variations. In contrast, our approach focuses on "abstracting" reasoning problems. This not only helps counteract distribution shifts but also facilitates the connection to symbolic tools for deriving solutions. We find that this abstraction process is better acquired through reinforcement learning (RL) than just supervised fine-tuning, which often fails to produce faithful abstractions. Our method, AbstraL -- which promotes abstract reasoning in LLMs using RL on granular abstraction data -- significantly mitigates performance degradation on recent GSM perturbation benchmarks.
Traditional Chinese Medicine (TCM) represents a rich repository of ancient medical knowledge that continues to play an important role in modern healthcare. Due to the complexity and breadth of the TCM literature, the integration of AI technologies is critical for its modernization and broader accessibility. However, this integration poses considerable challenges, including the interpretation of obscure classical Chinese texts and the modeling of intricate semantic relationships among TCM concepts. In this paper, we develop OpenTCM, an LLM-based system that combines a domain-specific TCM knowledge graph and Graph-based Retrieval-Augmented Generation (GraphRAG). First, we extract more than 3.73 million classical Chinese characters from 68 gynecological books in the Chinese Medical Classics Database, with the help of TCM and gynecology experts. Second, we construct a comprehensive multi-relational knowledge graph comprising more than 48,000 entities and 152,000 interrelationships, using customized prompts and Chinese-oriented LLMs such as DeepSeek and Kimi to ensure high-fidelity semantic understanding. Last, we integrate OpenTCM with this knowledge graph, enabling high-fidelity ingredient knowledge retrieval and diagnostic question-answering without model fine-tuning. Experimental evaluations demonstrate that OpenTCM achieves mean expert scores (MES) of 4.378 in ingredient information retrieval and 4.045 in diagnostic question-answering tasks, outperforming state-of-the-art solutions in real-world TCM use cases.
Large Language Models (LLMs) continue to exhibit vulnerabilities despite deliberate safety alignment efforts, posing significant risks to users and society. To safeguard against the risk of policy-violating content, system-level moderation via external guard models-designed to monitor LLM inputs and outputs and block potentially harmful content-has emerged as a prevalent mitigation strategy. Existing approaches of training guard models rely heavily on extensive human curated datasets and struggle with out-of-distribution threats, such as emerging harmful categories or jailbreak attacks. To address these limitations, we propose RSafe, an adaptive reasoning-based safeguard that conducts guided safety reasoning to provide robust protection within the scope of specified safety policies. RSafe operates in two stages: 1) guided reasoning, where it analyzes safety risks of input content through policy-guided step-by-step reasoning, and 2) reinforced alignment, where rule-based RL optimizes its reasoning paths to align with accurate safety prediction. This two-stage training paradigm enables RSafe to internalize safety principles to generalize safety protection capability over unseen or adversarial safety violation scenarios. During inference, RSafe accepts user-specified safety policies to provide enhanced safeguards tailored to specific safety requirements.
Query rewrite transforms SQL queries into semantically equivalent forms that run more efficiently. Existing approaches mainly rely on predefined rewrite rules, but they handle a limited subset of queries and can cause performance regressions. This limitation stems from three challenges of rule-based query rewrite: (1) it is hard to discover and verify new rules, (2) fixed rewrite rules do not generalize to new query patterns, and (3) some rewrite techniques cannot be expressed as fixed rules. Motivated by the fact that human experts exhibit significantly better rewrite ability but suffer from scalability, and Large Language Models (LLMs) have demonstrated nearly human-level semantic and reasoning abilities, we propose a new approach of using LLMs to rewrite SQL queries beyond rules. Due to the hallucination problems in LLMs, directly applying LLMs often leads to nonequivalent and suboptimal queries. To address this issue, we propose QUITE (query rewrite), a training-free and feedback-aware system based on LLM agents that rewrites SQL queries into semantically equivalent forms with significantly better performance, covering a broader range of query patterns and rewrite strategies compared to rule-based methods. Firstly, we design a multi-agent framework controlled by a finite state machine (FSM) to equip LLMs with the ability to use external tools and enhance the rewrite process with real-time database feedback. Secondly, we develop a rewrite middleware to enhance the ability of LLMs to generate optimized query equivalents. Finally, we employ a novel hint injection technique to improve execution plans for rewritten queries. Extensive experiments show that QUITE reduces query execution time by up to 35.8% over state-of-the-art approaches and produces 24.1% more rewrites than prior methods, covering query cases that earlier systems did not handle.
Large language models (LLMs) have made significant advancements across various tasks, but their safety alignment remain a major concern. Exploring jailbreak prompts can expose LLMs' vulnerabilities and guide efforts to secure them. Existing methods primarily design sophisticated instructions for the LLM to follow, or rely on multiple iterations, which could hinder the performance and efficiency of jailbreaks. In this work, we propose a novel jailbreak paradigm, Simple Assistive Task Linkage (SATA), which can effectively circumvent LLM safeguards and elicit harmful responses. Specifically, SATA first masks harmful keywords within a malicious query to generate a relatively benign query containing one or multiple [MASK] special tokens. It then employs a simple assistive task such as a masked language model task or an element lookup by position task to encode the semantics of the masked keywords. Finally, SATA links the assistive task with the masked query to jointly perform the jailbreak. Extensive experiments show that SATA achieves state-of-the-art performance and outperforms baselines by a large margin. Specifically, on AdvBench dataset, with mask language model (MLM) assistive task, SATA achieves an overall attack success rate (ASR) of 85% and harmful score (HS) of 4.57, and with element lookup by position (ELP) assistive task, SATA attains an overall ASR of 76% and HS of 4.43.
Large language models (LLMs) have demonstrated impressive capabilities across various natural language processing (NLP) tasks in recent years. However, their susceptibility to jailbreaks and perturbations necessitates additional evaluations. Many LLMs are multilingual, but safety-related training data contains mainly high-resource languages like English. This can leave them vulnerable to perturbations in low-resource languages such as Polish. We show how surprisingly strong attacks can be cheaply created by altering just a few characters and using a small proxy model for word importance calculation. We find that these character and word-level attacks drastically alter the predictions of different LLMs, suggesting a potential vulnerability that can be used to circumvent their internal safety mechanisms. We validate our attack construction methodology on Polish, a low-resource language, and find potential vulnerabilities of LLMs in this language. Additionally, we show how it can be extended to other languages. We release the created datasets and code for further research.
While Large Language Models (LLMs) have shown significant potential in assisting peer review, current methods often struggle to generate thorough and insightful reviews while maintaining efficiency. In this paper, we propose TreeReview, a novel framework that models paper review as a hierarchical and bidirectional question-answering process. TreeReview first constructs a tree of review questions by recursively decomposing high-level questions into fine-grained sub-questions and then resolves the question tree by iteratively aggregating answers from leaf to root to get the final review. Crucially, we incorporate a dynamic question expansion mechanism to enable deeper probing by generating follow-up questions when needed. We construct a benchmark derived from ICLR and NeurIPS venues to evaluate our method on full review generation and actionable feedback comments generation tasks. Experimental results of both LLM-based and human evaluation show that TreeReview outperforms strong baselines in providing comprehensive, in-depth, and expert-aligned review feedback, while reducing LLM token usage by up to 80% compared to computationally intensive approaches. Our code and benchmark dataset are available at https://github.com/YuanChang98/tree-review.