llm - 2025_09
Navigation
- Part 1
- Part 2
- Part 3
- Part 4
- Part 5
- Part 6
- Part 7
- Part 8
- Part 9
- Part 10
- Part 11
- Part 12
- Part 13
- Part 14
- Part 15
- Part 16
Papers
Large language models (LLMs) show strong reasoning via chain-of-thought (CoT) prompting, but the process is opaque, which makes verification, debugging, and control difficult in high-stakes settings. We present Vis-CoT, a human-in-the-loop framework that converts linear CoT text into an interactive reasoning graph. Users can visualize the logical flow, identify flawed steps, and intervene by pruning incorrect paths and grafting new, user-defined premises. This shifts interaction from passive observation to active collaboration, steering models toward more accurate and trustworthy conclusions. Across GSM8K and StrategyQA, Vis-CoT improves final-answer accuracy by up to 24 percentage points over non-interactive baselines. A user study also shows large gains in perceived usability and trust. Vis-CoT points to a practical path for more reliable, understandable, and collaborative reasoning by combining LLMs with targeted human oversight.
Large language models (LLMs) demonstrate remarkable performance on math word problems, yet they have been shown to struggle with meta-reasoning tasks such as identifying errors in student solutions. In this work, we investigate the challenge of locating the first error step in stepwise solutions using two error reasoning datasets: VtG and PRM800K. Our experiments show that state-of-the-art LLMs struggle to locate the first error step in student solutions even when given access to the reference solution. To that end, we propose an approach that generates an intermediate corrected student solution, aligning more closely with the original student's solution, which helps improve performance.
This paper proposes a reinforcement learning framework that employs Proximal Policy Optimization (PPO) to dynamically optimize the weights of multiple large language model (LLM)-generated formulaic alphas for stock trading strategies. Formulaic alphas are mathematically defined trading signals derived from price, volume, sentiment, and other data. Although recent studies have shown that LLMs can generate diverse and effective alphas, a critical challenge lies in how to adaptively integrate them under varying market conditions. To address this gap, we leverage the deepseek-r1-distill-llama-70b model to generate fifty alphas for five major stocks: Apple, HSBC, Pepsi, Toyota, and Tencent, and then use PPO to adjust their weights in real time. Experimental results demonstrate that the PPO-optimized strategy achieves strong returns and high Sharpe ratios across most stocks, outperforming both an equal-weighted alpha portfolio and traditional benchmarks such as the Nikkei 225, S&P 500, and Hang Seng Index. The findings highlight the importance of reinforcement learning in the allocation of alpha weights and show the potential of combining LLM-generated signals with adaptive optimization for robust financial forecasting and trading.
Current open-source training pipelines for Chinese medical language models predominantly emphasize optimizing training methodologies to enhance the performance of large language models (LLMs), yet lack comprehensive exploration into training data processing. To address this gap, we propose DPF-CM, a holistic Data Processing Framework for Chinese Medical LLMs training and deployment. DPF-CM comprises two core modules. The first module is a data processing pipeline tailored for model training. Beyond standard data processing operations, we (1) introduce a chained examples context-learning strategy to generate question-oriented instructions to mitigate the lack of instruction content, and (2) implement an ensemble-based filtering mechanism for preference data curation that averages multiple reward models to suppress noisy samples. The second module focuses on privacy preservation during model deployment. To prevent privacy risks from the inadvertent exposure of training data, we propose a Privacy Preserving Vector Database (PPVD) approach, which involves model memory search, high-risk database construction, secure database construction, and match-and-replace, four key stages to minimize privacy leakage during inference collectively. Experimental results show that DPF-CM significantly improves model accuracy, enabling our trained Chinese medical LLM to achieve state-of-the-art performance among open-source counterparts. Moreover, the framework reduces training data privacy leakage by 27%.
Understanding human intents from multimodal signals is critical for analyzing human behaviors and enhancing human-machine interactions in real-world scenarios. However, existing methods exhibit limitations in their modality-level reliance, constraining relational reasoning over fine-grained semantics for complex intent understanding. This paper proposes a novel LLM-Guided Semantic Relational Reasoning (LGSRR) method, which harnesses the expansive knowledge of large language models (LLMs) to establish semantic foundations that boost smaller models' relational reasoning performance. Specifically, an LLM-based strategy is proposed to extract fine-grained semantics as guidance for subsequent reasoning, driven by a shallow-to-deep Chain-of-Thought (CoT) that autonomously uncovers, describes, and ranks semantic cues by their importance without relying on manually defined priors. Besides, we formally model three fundamental types of semantic relations grounded in logical principles and analyze their nuanced interplay to enable more effective relational reasoning. Extensive experiments on multimodal intent and dialogue act recognition tasks demonstrate LGSRR's superiority over state-of-the-art methods, with consistent performance gains across diverse semantic understanding scenarios. The complete data and code are available at https://github.com/thuiar/LGSRR.
Large Language Models (LLMs), being generic task solvers, are versatile. However, despite the vast amount of data they are trained on, there are speculations about their adaptation capabilities to a new domain. Additionally, the simple fine-tuning of the model to incorporate knowledge of a new domain is computationally expensive and time-consuming. This becomes more challenging when the domain in question is also low-resource, and labeled data is unavailable. We leverage parameter-efficient fine-tuning techniques (PEFTs) on high-resource datasets to address these challenges to improve performance on unseen low-resource domains. Throughout our experiments, we evaluate whether intrinsic linguistic commonalities between datasets can be leveraged for efficient domain adaptation. We benchmark six PEFTs with \texttt{Llama-3-8B-Instruct} on 14 training datasets from the Scientific, Medical, Legal, and News domains for a Text Summarization task. Our experiments show that for low-resource domains, inference using Within-Domain Adapters can achieve better performance than Few-Shot as well as a much larger \texttt{Llama-3-70B-Instruct}. Lastly, in the absence of Within-Domain Adapters, we explore the concept of using Cross-Domain Adapters as well as the strategic combinations of adapters to leverage intrinsic language similarities across domains, facilitating better adaptability and performance in low-resource settings.
With the rapid advancement of artificial intelligence (AI), the proliferation of AI-generated content (AIGC) tasks has significantly accelerated developments in text-to-video generation. As a result, the field of video production is undergoing a transformative shift. However, conventional text-to-video models are typically constrained by high computational costs. In this study, we propose Video-Generation-Team (VGTeam), a novel slide show video generation system designed to redefine the video creation pipeline through the integration of large language models (LLMs). VGTeam is composed of a suite of communicative agents, each responsible for a distinct aspect of video generation, such as scriptwriting, scene creation, and audio design. These agents operate collaboratively within a chat tower workflow, transforming user-provided textual prompts into coherent, slide-style narrative videos. By emulating the sequential stages of traditional video production, VGTeam achieves remarkable improvements in both efficiency and scalability, while substantially reducing computational overhead. On average, the system generates videos at a cost of only $0.103, with a successful generation rate of 98.4%. Importantly, this framework maintains a high degree of creative fidelity and customization. The implications of VGTeam are far-reaching. It democratizes video production by enabling broader access to high-quality content creation without the need for extensive resources. Furthermore, it highlights the transformative potential of language models in creative domains and positions VGTeam as a pioneering system for next-generation content creation.
LLMs face significant challenges in systematic generalization, particularly when dealing with reasoning tasks requiring compositional rules and handling out-of-distribution examples. To address these challenges, we introduce an in-context learning methodology that improves the generalization capabilities of general purpose LLMs. Our approach employs an iterative example selection strategy, which incrementally constructs a tailored set of few-shot examples optimized to enhance model's performance on a given task. As a proof of concept, we apply this methodology to the resolution of algebraic expressions involving non-standard simplification rules, according to which the priority of addition and multiplication is changed. Our findings indicate that LLMs exhibit limited proficiency in these mathematical tasks. We further demonstrate that LLMs reasoning benefits from our iterative shot selection prompting strategy integrated with explicit reasoning instructions. Crucially, our experiments reveal that some LLMs achieve better generalization performances when prompted with simpler few-shot examples rather than complex ones following the test data distribution.
Quantization is a critical technique for accelerating LLM inference by reducing memory footprint and improving computational efficiency. Among various schemes, 4-bit weight and 8-bit activation quantization (W4A8) offers a strong balance between accuracy and performance. However, existing W4A8 GEMM kernels fall short in practice due to inefficient dequantization on CUDA Cores, which cannot keep pace with the high throughput of Tensor Cores. In this paper, we present LiquidGEMM, a hardware-efficient W4A8 GEMM kernel for efficient LLM serving. LiquidGEMM designs two key techniques: LiquidQuant, a hardware-efficient quantization method that enables fast, overflow-safe dequantization using just two arithmetic instructions per four elements; and an implicit fine-grained pipeline that fully overlaps weight loading, dequantization, and MMA across warp groups without software synchronization or redundant memory traffic. Experimental results show that LiquidGEMM achieves up to 2.90x speedup over state-of-the-art W4A8 kernels and up to 4.94x end-to-end system-level speedup. Compared to various quantized GEMM kernels in NVIDIA TensorRT-LLM, LiquidGEMM delivers 1.12-1.63x performance gains, and achieves up to 1.63x system-level speedup.
Hallucinations in Large Language Model (LLM) outputs for Question Answering (QA) tasks critically undermine their real-world reliability. This paper introduces an applied methodology for robust, one-shot hallucination detection, specifically designed for scenarios with limited data access, such as interacting with black-box LLM APIs that typically expose only a few top candidate log-probabilities per token. Our approach derives uncertainty indicators directly from these readily available log-probabilities generated during non-greedy decoding. We first derive an Entropy Production Rate (EPR) metric that offers baseline performance, later augmented with supervised learning. Our learned model uses features representing the entropic contributions of the accessible top-ranked tokens within a single generated sequence, requiring no multiple query re-runs. Evaluated across diverse QA datasets and multiple LLMs, this estimator significantly improves hallucination detection over using EPR alone. Crucially, high performance is demonstrated using only the typically small set of available log-probabilities (e.g., top <10 per token), confirming its practical efficiency and suitability for these API-constrained deployments. This work provides a readily deployable technique to enhance the trustworthiness of LLM responses from a single generation pass in QA and Retrieval-Augmented Generation (RAG) systems, with its utility further demonstrated in a finance framework analyzing responses to queries on annual reports from an industrial dataset.
With the proliferation of applications built upon LLM-driven multi-agent systems (MAS), the security of Web links has become a critical concern in ensuring system reliability. Once an agent is induced to visit a malicious website, attackers can use it as a springboard to conduct diverse subsequent attacks, which will drastically expand the attack surface. In this paper, we propose Web Fraud Attacks, a novel type of attack aiming at inducing MAS to visit malicious websites. We design 11 representative attack variants that encompass domain name tampering (homoglyph deception, character substitution, etc.), link structure camouflage (sub-directory nesting, sub-domain grafting, parameter obfuscation, etc.), and other deceptive techniques tailored to exploit MAS's vulnerabilities in link validation. Through extensive experiments on these crafted attack vectors, we demonstrate that Web fraud attacks not only exhibit significant destructive potential across different MAS architectures but also possess a distinct advantage in evasion: they circumvent the need for complex input formats such as jailbreaking, which inherently carry higher exposure risks. These results underscore the importance of addressing Web fraud attacks in LLM-driven MAS, as their stealthiness and destructiveness pose non-negligible threats to system security and user safety.
Large language models (LLMs) struggle to navigate culturally specific communication norms, limiting their effectiveness in global contexts. We focus on Persian taarof, a social norm in Iranian interactions, which is a sophisticated system of ritual politeness that emphasizes deference, modesty, and indirectness, yet remains absent from existing cultural benchmarks. We introduce TaarofBench, the first benchmark for evaluating LLM understanding of taarof, comprising 450 role-play scenarios covering 12 common social interaction topics, validated by native speakers. Our evaluation of five frontier LLMs reveals substantial gaps in cultural competence, with accuracy rates 40-48% below native speakers when taarof is culturally appropriate. Performance varies between interaction topics, improves with Persian-language prompts, and exhibits gender-based asymmetries. We also show that responses rated "polite" by standard metrics often violate taarof norms, indicating the limitations of Western politeness frameworks. Through supervised fine-tuning and Direct Preference Optimization, we achieve 21.8% and 42.3% improvement in model alignment with cultural expectations. Our human study with 33 participants (11 native Persian, 11 heritage, and 11 non-Iranian speakers) forms baselines in varying degrees of familiarity with Persian norms. This work lays the foundation for developing diverse and culturally aware LLMs, enabling applications that better navigate complex social interactions.
While robots have previously utilized rule-based systems or probabilistic models for user interaction, the rapid evolution of large language models (LLMs) presents new opportunities to develop LLM-powered robots for enhanced human-robot interaction (HRI). To fully realize these capabilities, however, robots need to collect data such as audio, fine-grained images, video, and locations. As a result, LLMs often process sensitive personal information, particularly within private environments, such as homes. Given the tension between utility and privacy risks, evaluating how current LLMs manage sensitive data is critical. Specifically, we aim to explore the extent to which out-of-the-box LLMs are privacy-aware in the context of household robots. In this work, we present a set of privacy-relevant scenarios developed using the Contextual Integrity (CI) framework. We first surveyed users' privacy preferences regarding in-home robot behaviors and then examined how their privacy orientations affected their choices of these behaviors (N = 450). We then provided the same set of scenarios and questions to state-of-the-art LLMs (N = 10) and found that the agreement between humans and LLMs was generally low. To further investigate the capabilities of LLMs as potential privacy controllers, we implemented four additional prompting strategies and compared their results. We discuss the performance of the evaluated models as well as the implications and potential of AI privacy awareness in human-robot interaction.
We present ExCyTIn-Bench, the first benchmark to Evaluate an LLM agent x on the task of Cyber Threat Investigation through security questions derived from investigation graphs. Real-world security analysts must sift through a large number of heterogeneous alert signals and security logs, follow multi-hop chains of evidence, and compile an incident report. With the developments of LLMs, building LLM-based agents for automatic thread investigation is a promising direction. To assist the development and evaluation of LLM agents, we construct a dataset from a controlled Azure tenant that covers 8 simulated real-world multi-step attacks, 57 log tables from Microsoft Sentinel and related services, and 589 automatically generated questions. We leverage security logs extracted with expert-crafted detection logic to build threat investigation graphs, and then generate questions with LLMs using paired nodes on the graph, taking the start node as background context and the end node as answer. Anchoring each question to these explicit nodes and edges not only provides automatic, explainable ground truth answers but also makes the pipeline reusable and readily extensible to new logs. This also enables the automatic generation of procedural tasks with verifiable rewards, which can be naturally extended to training agents via reinforcement learning. Our comprehensive experiments with different models confirm the difficulty of the task: with the base setting, the average reward across all evaluated models is 0.249, and the best achieved is 0.368, leaving substantial headroom for future research. Code and data are coming soon!
We present TRACE-CS, a novel hybrid system that combines symbolic reasoning with large language models (LLMs)to address contrastive queries in course scheduling problems. TRACE-CS leverages logic-based techniques to encode scheduling constraints and generate provably correct explanations, while utilizing an LLM to process natural language queries and refine logical explanations into user friendly responses. This system showcases how combining symbolic KR methods with LLMs creates explainable AI agents that balance logical correctness with natural language accessibility, addressing a fundamental challenge in deployed scheduling systems.
Intelligent control of Unmanned Aerial Vehicles (UAVs) swarms has emerged as a critical research focus, and it typically requires the swarm to navigate effectively while avoiding obstacles and achieving continuous coverage over multiple mission targets. Although traditional Multi-Agent Reinforcement Learning (MARL) approaches offer dynamic adaptability, they are hindered by the semantic gap in numerical communication and the rigidity of homogeneous role structures, resulting in poor generalization and limited task scalability. Recent advances in Large Language Model (LLM)-based control frameworks demonstrate strong semantic reasoning capabilities by leveraging extensive prior knowledge. However, due to the lack of online learning and over-reliance on static priors, these works often struggle with effective exploration, leading to reduced individual potential and overall system performance. To address these limitations, we propose a Role-Adaptive LLM-Driven Yoked navigation algorithm RALLY. Specifically, we first develop an LLM-driven semantic decision framework that uses structured natural language for efficient semantic communication and collaborative reasoning. Afterward, we introduce a dynamic role-heterogeneity mechanism for adaptive role switching and personalized decision-making. Furthermore, we propose a Role-value Mixing Network (RMIX)-based assignment strategy that integrates LLM offline priors with MARL online policies to enable semi-offline training of role selection strategies. Experiments in the Multi-Agent Particle Environment (MPE) environment and a Software-In-The-Loop (SITL) platform demonstrate that RALLY outperforms conventional approaches in terms of task coverage, convergence speed, and generalization, highlighting its strong potential for collaborative navigation in agentic multi-UAV systems.