LLM-as-Judge has emerged as a scalable alternative to human evaluation, enabling large language models (LLMs) to provide reward signals in trainings. While recent work has explored multi-agent extensions such as multi-agent debate and meta-judging to enhance evaluation quality, the question of how intrinsic biases manifest in these settings remains underexplored. In this study, we conduct a systematic analysis of four diverse bias types: position bias, verbosity bias, chain-of-thought bias, and bandwagon bias. We evaluate these biases across two widely adopted multi-agent LLM-as-Judge frameworks: Multi-Agent-Debate and LLM-as-Meta-Judge. Our results show that debate framework amplifies biases sharply after the initial debate, and this increased bias is sustained in subsequent rounds, while meta-judge approaches exhibit greater resistance. We further investigate the incorporation of PINE, a leading single-agent debiasing method, as a bias-free agent within these systems. The results reveal that this bias-free agent effectively reduces biases in debate settings but provides less benefit in meta-judge scenarios. Our work provides a comprehensive study of bias behavior in multi-agent LLM-as-Judge systems and highlights the need for targeted bias mitigation strategies in collaborative evaluation settings.

The recent trend of self-sovereign Decentralized AI Agents (DeAgents) combines Large Language Model (LLM)-based AI agents with decentralization technologies such as blockchain smart contracts and trusted execution environments (TEEs). These tamper-resistant trustless substrates allow agents to achieve self-sovereignty through ownership of cryptowallet private keys and control of digital assets and social media accounts. DeAgents eliminate centralized control and reduce human intervention, addressing key trust concerns inherent in centralized AI systems. This contributes to social computing by enabling new human cooperative paradigm "intelligence as commons." However, given ongoing challenges in LLM reliability such as hallucinations, this creates paradoxical tension between trustlessness and unreliable autonomy. This study addresses this empirical research gap through interviews with DeAgents stakeholders-experts, founders, and developers-to examine their motivations, benefits, and governance dilemmas. The findings will guide future DeAgents system and protocol design and inform discussions about governance in sociotechnical AI systems in the future agentic web.

Large Language Models (LLMs) often fail on complex reasoning tasks due to flawed question comprehension, not just flawed logic. This paper presents a systematic investigation into these comprehension failures. Our work yields three key insights: (1) the step-by-step principle, effective for calculation, can be migrated to the reading process to enhance comprehension; (2) increasing the proportion of question-related tokens (e.g., via repetition) succeeds by refocusing attention, a mechanism that can be explicitly controlled; and (3) backward dependencies represent a core bottleneck for decoder-only models that persists even with strong methods like Chain-of-Thought. Based on these findings, we introduce the Step-by-Step Reading (SSR) family of prompts. This multi-stage approach culminates in SSR++, a method specifically engineered to deepen model comprehension by guiding it to parse questions with finer granularity, focus attention on critical tokens, and resolve backward dependencies through iterative re-contextualization. SSR++ sets a new state-of-the-art on multiple reasoning benchmarks, and our analysis confirms it works by directly mitigating semantic misunderstanding. These results demonstrate that guiding how a model reads is a powerful and efficient method for improving its reasoning ability.

The problem of data contamination is now almost inevitable during the development of large language models (LLMs), with the training data commonly integrating those evaluation benchmarks even unintentionally. This problem subsequently makes it hard to benchmark LLMs fairly. Instead of constructing contamination-free datasets (quite hard), we propose a novel framework, \textbf{LNE-Blocking}, to restore model performance prior to contamination on potentially leaked datasets. Our framework consists of two components: contamination detection and disruption operation. For the prompt, the framework first uses the contamination detection method, \textbf{LNE}, to assess the extent of contamination in the model. Based on this, it adjusts the intensity of the disruption operation, \textbf{Blocking}, to elicit non-memorized responses from the model. Our framework is the first to efficiently restore the model's greedy decoding performance. This comes with a strong performance on multiple datasets with potential leakage risks, and it consistently achieves stable recovery results across different models and varying levels of data contamination. We release the code at https://github.com/RuijieH/LNE-Blocking to facilitate research.

Multimodal large language models have various practical applications that demand strong reasoning abilities. Despite recent advancements, these models still struggle to solve complex geometric problems. A key challenge stems from the lack of high-quality image-text pair datasets for understanding geometric images. Furthermore, most template-based data synthesis pipelines typically fail to generalize to questions beyond their predefined templates. In this paper, we bridge this gap by introducing a complementary process of Reinforcement Learning with Verifiable Rewards (RLVR) into the data generation pipeline. By adopting RLVR to refine captions for geometric images synthesized from 50 basic geometric relations and using reward signals derived from mathematical problem-solving tasks, our pipeline successfully captures the key features of geometry problem-solving. This enables better task generalization and yields non-trivial improvements. Furthermore, even in out-of-distribution scenarios, the generated dataset enhances the general reasoning capabilities of multimodal large language models, yielding accuracy improvements of $2.8\%\text{-}4.8\%$ in statistics, arithmetic, algebraic, and numerical tasks with non-geometric input images of MathVista and MathVerse, along with $2.4\%\text{-}3.9\%$ improvements in Art, Design, Tech, and Engineering tasks in MMMU.

Slide decks, serving as digital reports that bridge the gap between presentation slides and written documents, are a prevalent medium for conveying information in both academic and corporate settings. Their multimodal nature, combining text, images, and charts, presents challenges for retrieval-augmented generation systems, where the quality of retrieval directly impacts downstream performance. Traditional approaches to slide retrieval often involve separate indexing of modalities, which can increase complexity and lose contextual information. This paper investigates various methodologies for effective slide retrieval, including visual late-interaction embedding models like ColPali, the use of visual rerankers, and hybrid retrieval techniques that combine dense retrieval with BM25, further enhanced by textual rerankers and fusion methods like Reciprocal Rank Fusion. A novel Vision-Language Models-based captioning pipeline is also evaluated, demonstrating significantly reduced embedding storage requirements compared to visual late-interaction techniques, alongside comparable retrieval performance. Our analysis extends to the practical aspects of these methods, evaluating their runtime performance and storage demands alongside retrieval efficacy, thus offering practical guidance for the selection and development of efficient and robust slide retrieval systems for real-world applications.

High memory demands of generative language models have drawn attention to quantization, which reduces computational cost, memory usage, and latency by mapping model weights to lower-precision integers. Approaches such as GPTQ effectively minimize input-weight product errors during quantization; however, recent empirical studies show that they can increase biased outputs and degrade performance on fairness benchmarks, and it remains unclear which specific weights cause this issue. In this work, we draw new links between quantization and model fairness by adding explicit group-fairness constraints to the quantization objective and introduce Fair-GPTQ, the first quantization method explicitly designed to reduce unfairness in large language models. The added constraints guide the learning of the rounding operation toward less-biased text generation for protected groups. Specifically, we focus on stereotype generation involving occupational bias and discriminatory language spanning gender, race, and religion. Fair-GPTQ has minimal impact on performance, preserving at least 90% of baseline accuracy on zero-shot benchmarks, reduces unfairness relative to a half-precision model, and retains the memory and speed benefits of 4-bit quantization. We also compare the performance of Fair-GPTQ with existing debiasing methods and find that it achieves performance on par with the iterative null-space projection debiasing approach on racial-stereotype benchmarks. Overall, the results validate our theoretical solution to the quantization problem with a group-bias term, highlight its applicability for reducing group bias at quantization time in generative models, and demonstrate that our approach can further be used to analyze channel- and weight-level contributions to fairness during quantization.

Fuzz testing is one of the most effective techniques for finding software vulnerabilities. While modern fuzzers can generate inputs and monitor executions automatically, the overall workflow, from analyzing a codebase, to configuring harnesses, to triaging results, still requires substantial manual effort. Prior attempts focused on single stages such as harness synthesis or input minimization, leaving researchers to manually connect the pieces into a complete fuzzing campaign. We introduce Orion, a framework that automates the the manual bottlenecks of fuzzing by integrating LLM reasoning with traditional tools, allowing campaigns to scale to settings where human effort alone was impractical. Orion uses LLMs for code reasoning and semantic guidance, while relying on deterministic tools for verification, iterative refinement, and tasks that require precision. Across our benchmark suite, Orion reduces human effort by 46-204x depending on the workflow stage, and we demonstrate its effectiveness through the discovery of two previously unknown vulnerabilities in the widely used open-source clib library.

Retrieval-Augmented Generation (RAG) enhances large language models (LLMs) by retrieving relevant documents from external sources to improve factual accuracy and verifiability. However, this reliance introduces new attack surfaces within the retrieval pipeline, beyond the LLM itself. While prior RAG attacks have exposed such vulnerabilities, they largely rely on manipulating user queries, which is often infeasible in practice due to fixed or protected user inputs. This narrow focus overlooks a more realistic and stealthy vector: instructional prompts, which are widely reused, publicly shared, and rarely audited. Their implicit trust makes them a compelling target for adversaries to manipulate RAG behavior covertly. We introduce a novel attack for Adversarial Instructional Prompt (AIP) that exploits adversarial instructional prompts to manipulate RAG outputs by subtly altering retrieval behavior. By shifting the attack surface to the instructional prompts, AIP reveals how trusted yet seemingly benign interface components can be weaponized to degrade system integrity. The attack is crafted to achieve three goals: (1) naturalness, to evade user detection; (2) utility, to encourage use of prompts; and (3) robustness, to remain effective across diverse query variations. We propose a diverse query generation strategy that simulates realistic linguistic variation in user queries, enabling the discovery of prompts that generalize across paraphrases and rephrasings. Building on this, a genetic algorithm-based joint optimization is developed to evolve adversarial prompts by balancing attack success, clean-task utility, and stealthiness. Experimental results show that AIP achieves up to 95.23% ASR while preserving benign functionality. These findings uncover a critical and previously overlooked vulnerability in RAG systems, emphasizing the need to reassess the shared instructional prompts.

Foundation models trained on web-scale data have revolutionized robotics, but their application to low-level control remains largely limited to behavioral cloning. Drawing inspiration from the success of the reinforcement learning stage in fine-tuning large language models, we propose a two-stage post-training approach for robotics. The first stage, Supervised Fine-Tuning (SFT), fine-tunes pretrained foundation models using both: a) behavioral cloning, and b) steps-to-go prediction objectives. In the second stage, Self-Improvement, steps-to-go prediction enables the extraction of a well-shaped reward function and a robust success detector, enabling a fleet of robots to autonomously practice downstream tasks with minimal human supervision. Through extensive experiments on real-world and simulated robot embodiments, our novel post-training recipe unveils significant results on Embodied Foundation Models. First, we demonstrate that the combination of SFT and Self-Improvement is significantly more sample-efficient than scaling imitation data collection for supervised learning, and that it leads to policies with significantly higher success rates. Further ablations highlight that the combination of web-scale pretraining and Self-Improvement is the key to this sample-efficiency. Next, we demonstrate that our proposed combination uniquely unlocks a capability that current methods cannot achieve: autonomously practicing and acquiring novel skills that generalize far beyond the behaviors observed in the imitation learning datasets used during training. These findings highlight the transformative potential of combining pretrained foundation models with online Self-Improvement to enable autonomous skill acquisition in robotics. Our project website can be found at https://self-improving-efms.github.io .