Browse by signal

RoboTic-Tac-Toe is an interactive game where two physical robots move around a tic-tac-toe board, with both the gameplay and robots’ movements orchestrated by LLMs. Players can control the robots using natural language commands, directing them to place their markers on the game board. In this post, we explore the architecture and prompt engineering techniques used to reason about a tic-tac-toe game and decide the next best game strategy and movement plan for the current player.

In what appeared to be a bid to soak up some of Google's limelight prior to the launch of its new Gemini 3 flagship AI model — now recorded as the most powerful LLM in the world by multiple independent evaluators — Elon Musk's rival AI startup xAI last night unveiled its newest large language model, Grok 4.1. The model is now live for consumer use on Grok.com, social network X (formerly Twitter), and the company’s iOS and Android mobile apps, and it arrives with major architectural and usability enhancements, among them: faster reasoning, improved emotional intelligence, and significantly reduced hallucination rates. xAI also commendably published a white paper on its evaluations and including a small bit on training process here . Across public benchmarks, Grok 4.1 has vaulted to the top of the leaderboard, outperforming rival models from Anthropic, OpenAI, and Google — at least, Google's pre-Gemini 3 model (Gemini 2.5 Pro). It builds upon the success of xAI's Grok-4 Fast, which VentureBeat covered favorably shortly following its release back in September 2025. However, enterprise developers looking to integrate the new and improved model Grok 4.1 into production environments will find one major constraint: it's not yet available through xAI’s public API . Despite its high benchmarks, Grok 4.1 remains confined to xAI’s consumer-facing interfaces, with no announced timeline for API exposure. At present, only older models—including Grok 4 Fast (reasoning and non-reasoning variants), Grok 4 0709, and legacy models such as Grok 3, Grok 3 Mini, and Grok 2 Vision—are available for programmatic use via the xAI developer API. These support up to 2 million tokens of context, with token pricing ranging from $0.20 to $3.00 per million depending on the configuration. For now, this limits Grok 4.1’s utility in enterprise workflows that rely on backend integration, fine-tuned agentic pipelines, or scalable internal tooling. While the consumer rollout positions Grok 4.1 as the most capable LLM in xAI’s portfolio, production deployments in enterprise environments remain on hold. Model Design and Deployment Strategy Grok 4.1 arrives in two configurations: a fast-response, low-latency mode for immediate replies, and a “thinking” mode that engages in multi-step reasoning before producing output. Both versions are live for end users and are selectable via the model picker in xAI’s apps. The two configurations differ not just in latency but also in how deeply the model processes prompts. Grok 4.1 Thinking leverages internal planning and deliberation mechanisms, while the standard version prioritizes speed. Despite the difference in architecture, both scored higher than any competing models in blind preference and benchmark testing. Leading the Field in Human and Expert Evaluation On the LMArena Text Arena leaderboard , Grok 4.1 Thinking briefly held the top position with a normalized Elo score of 1483 — then was dethroned a few hours later with Google's release of Gemini 3 and its incredible 1501 Elo score. The non-thinking version of Grok 4.1 also fares well on the index, however, at 1465. These scores place Grok 4.1 above Google’s Gemini 2.5 Pro, Anthropic’s Claude 4.5 series, and OpenAI’s GPT-4.5 preview. In creative writing, Grok 4.1 ranks second only to Polaris Alpha (an early GPT-5.1 variant), with the “thinking” model earning a score of 1721.9 on the Creative Writing v3 benchmark. This marks a roughly 600-point improvement over previous Grok iterations. Similarly, in the Arena Expert leaderboard, which aggregates feedback from professional reviewers, Grok 4.1 Thinking again leads the field with a score of 1510. The gains are especially notable given that Grok 4.1 was released only two months after Grok 4 Fast, highlighting the accelerated development pace at xAI. Core Improvements Over Previous Generations Technically, Grok 4.1 represents a significant leap in real-world usability. Visual capabilities—previously limited in Grok 4—have been upgraded to enable robust image and video understanding, including chart analysis and OCR-level text extraction. Multimodal reliability was a pain point in prior versions and has now been addressed. Token-level latency has been reduced by approximately 28 percent while preserving reasoning depth. In long-context tasks, Grok 4.1 maintains coherent output up to 1 million tokens, improving on Grok 4’s tendency to degrade past the 300,000 token mark. xAI has also improved the model's tool orchestration capabilities. Grok 4.1 can now plan and execute multiple external tools in parallel, reducing the number of interaction cycles required to complete multi-step queries. According to internal test logs, some research tasks that previously required four steps can now be completed in one or two. Other alignment improvements include better truth calibration—reducing the tendency to hedge or soften politically sensitive outputs—and more natural, human-like prosody in voice mode, with support for different speaking styles and accents. Safety and Adversarial Robustness As part of its risk management framework, xAI evaluated Grok 4.1 for refusal behavior, hallucination resistance, sycophancy, and dual-use safety. The hallucination rate in non-reasoning mode has dropped from 12.09 percent in Grok 4 Fast to just 4.22 percent — a roughly 65% improvement. The model also scored 2.97 percent on FActScore, a factual QA benchmark, down from 9.89 percent in earlier versions. In the domain of adversarial robustness, Grok 4.1 has been tested with prompt injection attacks, jailbreak prompts, and sensitive chemistry and biology queries. Safety filters showed low false negative rates, especially for restricted chemical knowledge (0.00 percent) and restricted biological queries (0.03 percent). The model’s ability to resist manipulation in persuasion benchmarks, such as MakeMeSay, also appears strong—it registered a 0 percent success rate as an attacker. Limited Enterprise Access via API Despite these gains, Grok 4.1 remains unavailable to enterprise users through xAI’s API. According to the company’s public documentation , the latest available models for developers are Grok 4 Fast (both reasoning and non-reasoning variants), each supporting up to 2 million tokens of context at pricing tiers ranging from $0.20 to $0.50 per million tokens. These are backed by a 4M tokens-per-minute throughput limit and 480 requests per minute (RPM) rate cap. By contrast, Grok 4.1 is accessible only through xAI’s consumer-facing properties—X, Grok.com, and the mobile apps. This means organizations cannot yet deploy Grok 4.1 via fine-tuned internal workflows, multi-agent chains, or real-time product integrations. Industry Reception and Next Steps The release has been met with strong public and industry feedback. Elon Musk, founder of xAI, posted a brief endorsement, calling it “a great model” and congratulating the team. AI benchmark platforms have praised the leap in usability and linguistic nuance. For enterprise customers, however, the picture is more mixed. Grok 4.1’s performance represents a breakthrough for general-purpose and creative tasks, but until API access is enabled, it will remain a consumer-first product with limited enterprise applicability. As competitive models from OpenAI, Google, and Anthropic continue to evolve, xAI’s next strategic move may hinge on when—and how—it opens Grok 4.1 to external developers.

arXiv:2511.11252v1 Announce Type: new Abstract: Autonomous aerial systems increasingly rely on large language models (LLMs) for mission planning, perception, and decision-making, yet the lack of standardized and physically grounded benchmarks limits systematic evaluation of their reasoning capabilities. To address this gap, we introduce UAVBench, an open benchmark dataset comprising 50,000 validated UAV flight scenarios generated through taxonomy-guided LLM prompting and multi-stage safety validation. Each scenario is encoded in a structured JSON schema that includes mission objectives, vehicle configuration, environmental conditions, and quantitative risk labels, providing a unified representation of UAV operations across diverse domains. Building on this foundation, we present UAVBench_MCQ, a reasoning-oriented extension containing 50,000 multiple-choice questions spanning ten cognitive and ethical reasoning styles, ranging from aerodynamics and navigation to multi-agent coordination and integrated reasoning. This framework enables interpretable and machine-checkable assessment of UAV-specific cognition under realistic operational contexts. We evaluate 32 state-of-the-art LLMs, including GPT-5, ChatGPT-4o, Gemini 2.5 Flash, DeepSeek V3, Qwen3 235B, and ERNIE 4.5 300B, and find strong performance in perception and policy reasoning but persistent challenges in ethics-aware and resource-constrained decision-making. UAVBench establishes a reproducible and physically grounded foundation for benchmarking agentic AI in autonomous aerial systems and advancing next-generation UAV reasoning intelligence. To support open science and reproducibility, we release the UAVBench dataset, the UAVBench_MCQ benchmark, evaluation scripts, and all related materials on GitHub at https://github.com/maferrag/UAVBench

arXiv:2511.11182v1 Announce Type: new Abstract: Hallucination continues to pose a major obstacle in the reasoning capabilities of large language models (LLMs). Although the Multi-Agent Debate (MAD) paradigm offers a promising solution by promoting consensus among multiple agents to enhance reliability, it relies on the unrealistic assumption that all debaters are rational and reflective, which is a condition that may not hold when agents themselves are prone to hallucinations. To address this gap, we introduce the Multi-agent Undercover Gaming (MUG) protocol, inspired by social deduction games like "Who is Undercover?". MUG reframes MAD as a process of detecting "undercover" agents (those suffering from hallucinations) by employing multimodal counterfactual tests. Specifically, we modify reference images to introduce counterfactual evidence and observe whether agents can accurately identify these changes, providing ground-truth for identifying hallucinating agents and enabling robust, crowd-powered multimodal reasoning. MUG advances MAD protocols along three key dimensions: (1) enabling factual verification beyond statistical consensus through counterfactual testing; (2) introducing cross-evidence reasoning via dynamically modified evidence sources instead of relying on static inputs; and (3) fostering active reasoning, where agents engage in probing discussions rather than passively answering questions. Collectively, these innovations offer a more reliable and effective framework for multimodal reasoning in LLMs. The source code can be accessed at https://github.com/YongLD/MUG.git.

arXiv:2511.11040v1 Announce Type: new Abstract: Recent studies on LLM agent scaling have highlighted the potential of Multi-Agent Debate (MAD) to enhance reasoning abilities. However, the critical aspect of role allocation strategies remains underexplored. In this study, we demonstrate that allocating roles with differing viewpoints to specific positions significantly impacts MAD's performance in reasoning tasks. Specifically, we find a novel role allocation strategy, "Truth Last", which can improve MAD performance by up to 22% in reasoning tasks. To address the issue of unknown truth in practical applications, we propose the Multi-Agent Debate Consistency (MADC) strategy, which systematically simulates and optimizes its core mechanisms. MADC incorporates path consistency to assess agreement among independent roles, simulating the role with the highest consistency score as the truth. We validated MADC across a range of LLMs (9 models), including the DeepSeek-R1 Distilled Models, on challenging reasoning tasks. MADC consistently demonstrated advanced performance, effectively overcoming MAD's performance bottlenecks and providing a crucial pathway for further improvements in LLM agent scaling.

arXiv:2511.11017v1 Announce Type: new Abstract: The rapid expansion of e-commerce platforms generates vast amounts of unstructured product data, creating significant challenges for information retrieval, recommendation systems, and data analytics. Knowledge Graphs (KGs) offer a structured, interpretable format to organize such data, yet constructing product-specific KGs remains a complex and manual process. This paper introduces a fully automated, AI agent-driven framework for constructing product knowledge graphs directly from unstructured product descriptions. Leveraging Large Language Models (LLMs), our method operates in three stages using dedicated agents: ontology creation and expansion, ontology refinement, and knowledge graph population. This agent-based approach ensures semantic coherence, scalability, and high-quality output without relying on predefined schemas or handcrafted extraction rules. We evaluate the system on a real-world dataset of air conditioner product descriptions, demonstrating strong performance in both ontology generation and KG population. The framework achieves over 97\% property coverage and minimal redundancy, validating its effectiveness and practical applicability. Our work highlights the potential of LLMs to automate structured knowledge extraction in retail, providing a scalable path toward intelligent product data integration and utilization.

arXiv:2511.10890v1 Announce Type: new Abstract: Understanding the interactions between biomarkers among brain regions during neurodegenerative disease is essential for unravelling the mechanisms underlying disease progression. For example, pathophysiological models of Alzheimer's Disease (AD) typically describe how variables, such as regional levels of toxic proteins, interact spatiotemporally within a dynamical system driven by an underlying biological substrate, often based on brain connectivity. However, current methods grossly oversimplify the complex relationship between brain connectivity by assuming a single-modality brain connectome as the disease-spreading substrate. This leads to inaccurate predictions of pathology spread, especially during the long-term progression period. Meanhwile, other methods of learning such a graph in a purely data-driven way face the identifiability issue due to lack of proper constraint. We thus present a novel framework that uses Large Language Models (LLMs) as expert guides on the interaction of regional variables to enhance learning of disease progression from irregularly sampled longitudinal patient data. By leveraging LLMs' ability to synthesize multi-modal relationships and incorporate diverse disease-driving mechanisms, our method simultaneously optimizes 1) the construction of long-term disease trajectories from individual-level observations and 2) the biologically-constrained graph structure that captures interactions among brain regions with better identifiability. We demonstrate the new approach by estimating the pathology propagation using tau-PET imaging data from an Alzheimer's disease cohort. The new framework demonstrates superior prediction accuracy and interpretability compared to traditional approaches while revealing additional disease-driving factors beyond conventional connectivity measures.

arXiv:2511.10788v1 Announce Type: new Abstract: Recent advances in large language models (LLMs) have made reasoning a central benchmark for evaluating intelligence. While prior surveys focus on efficiency by examining how to shorten reasoning chains or reduce computation, this view overlooks a fundamental challenge: current LLMs apply uniform reasoning strategies regardless of task complexity, generating long traces for trivial problems while failing to extend reasoning for difficult tasks. This survey reframes reasoning through the lens of {adaptivity}: the capability to allocate reasoning effort based on input characteristics such as difficulty and uncertainty. We make three contributions. First, we formalize deductive, inductive, and abductive reasoning within the LLM context, connecting these classical cognitive paradigms with their algorithmic realizations. Second, we formalize adaptive reasoning as a control-augmented policy optimization problem balancing task performance with computational cost, distinguishing learned policies from inference-time control mechanisms. Third, we propose a systematic taxonomy organizing existing methods into training-based approaches that internalize adaptivity through reinforcement learning, supervised fine-tuning, and learned controllers, and training-free approaches that achieve adaptivity through prompt conditioning, feedback-driven halting, and modular composition. This framework clarifies how different mechanisms realize adaptive reasoning in practice and enables systematic comparison across diverse strategies. We conclude by identifying open challenges in self-evaluation, meta-reasoning, and human-aligned reasoning control.

arXiv:2511.11233v1 Announce Type: new Abstract: Table reasoning with the large language models (LLMs) is a fundamental path toward building intelligent systems that can understand and analyze over structured data. While recent progress has shown promising results, they still suffer from two key limitations: (i) the reasoning processes lack the depth and iterative refinement characteristic of human cognition; and (ii) the reasoning processes exhibit instability, which compromises their reliability in downstream applications. In this work, we present STaR (slow-thinking for table reasoning), a new framework achieving cognitive table reasoning, in which LLMs are equipped with slow-thinking capabilities by explicitly modeling step-by-step thinking and uncertainty-aware inference. During training, STaR employs two-stage difficulty-aware reinforcement learning (DRL), progressively learning from simple to complex queries under a composite reward. During inference, STaR performs trajectory-level uncertainty quantification by integrating token-level confidence and answer consistency, enabling selection of more credible reasoning paths. Extensive experiments on benchmarks demonstrate that STaR achieves superior performance and enhanced reasoning stability. Moreover, strong generalization over out-of-domain datasets further demonstrates STaR's potential as a reliable and cognitively inspired solution for table reasoning with LLMs.

arXiv:2511.10810v1 Announce Type: new Abstract: Operational safety at mission-critical work sites is a top priority given the complex and hazardous nature of daily tasks. This paper presents the Human-Agent Risk Navigation and Event Safety System (HARNESS), a modular AI framework designed to forecast hazardous events and analyze operational risks in U.S. Department of Energy (DOE) environments. HARNESS integrates Large Language Models (LLMs) with structured work data, historical event retrieval, and risk analysis to proactively identify potential hazards. A human-in-the-loop mechanism allows subject matter experts (SMEs) to refine predictions, creating an adaptive learning loop that enhances performance over time. By combining SME collaboration with iterative agentic reasoning, HARNESS improves the reliability and efficiency of predictive safety systems. Preliminary deployment shows promising results, with future work focusing on quantitative evaluation of accuracy, SME agreement, and decision latency reduction.

arXiv:2511.10687v1 Announce Type: cross Abstract: Large Language Models (LLMs) in multi-agent systems (MAS) have shown promise for complex tasks, yet current training methods lack principled ways to connect system-level evaluation with agent-level and message-level learning. We propose a theoretical framework that unifies cooperative game-theoretic attribution with process reward modeling to transform system evaluation into agent credit and then into response-level signals. Unlike prior approaches that rely only on attribution (e.g., Shapley) or step-level labels (e.g., PRM), our method produces local, signed, and credit-conserving signals. In success cases, Shapley-based credit assignment fairly allocates outcomes across agents and is refined into per-message rewards that promote cooperation while discouraging redundancy or sabotage. In failure cases, first-error localization yields repair-aware preferences that penalize harmful steps while rewarding corrective attempts. The resulting signals are bounded, cooperative, and directly compatible with reinforcement-based or preference-based post-training, providing a unified and auditable pathway from global evaluation to local supervision in LLM multi-agent training. Our contribution is conceptual: we present a theoretical foundation and training signals, leaving empirical validation for future work.

arXiv:2511.10676v1 Announce Type: cross Abstract: Mixture-of-Experts (MoE) Large Language Models (LLMs) efficiently scale-up the model while keeping relatively low inference cost. As MoE models only activate part of the experts, related work has proposed expert prediction and caching methods to prefetch the experts for faster inference. However, existing approaches utilize the activations from the previous layer for prediction, incurring low accuracy and leave the first layer unoptimized. Applying complex layers or even training standalone networks for better prediction introduces high computation overhead. In this paper, we propose pre-attention expert prediction to achieve accurate and lightweight expert prefetching. The key insight is that some functions in LLMs are ranking-preserving, indicating that matching the ranking of selected experts using simple linear functions is possible. Therefore, we utilize the activations before the attention block in the same layer with 2 linear functions and ranking-aware loss to achieve accurate prediction, which also supports prefetching in the first layer. Our lightweight, pre-attention expert routers achieve 93.03% accuracy on DeepSeek V2 Lite, 94.69% on Qwen3-30B, and 97.62% on Phi-mini-MoE, showing about 15% improvement on absolute accuracy over the state-of-the-art methods.

arXiv:2511.10674v1 Announce Type: cross Abstract: Large Language Models (LLMs) can generate SQL queries from natural language questions but struggle with database-specific schemas and tacit domain knowledge. We introduce a framework for continual learning from human feedback in text-to-SQL, where a learning agent receives natural language feedback to refine queries and distills the revealed knowledge for reuse on future tasks. This distilled knowledge is stored in a structured memory, enabling the agent to improve execution accuracy over time. We design and evaluate multiple variations of a learning agent architecture that vary in how they capture and retrieve past experiences. Experiments on the BIRD benchmark Dev set show that memory-augmented agents, particularly the Procedural Agent, achieve significant accuracy gains and error reduction by leveraging human-in-the-loop feedback. Our results highlight the importance of transforming tacit human expertise into reusable knowledge, paving the way for more adaptive, domain-aware text-to-SQL systems that continually learn from a human-in-the-loop.

arXiv:2511.10670v1 Announce Type: cross Abstract: Code-switching (CS) speech translation (ST) refers to translating speech that alternates between two or more languages into a target language text, which poses significant challenges due to the complexity of semantic modeling and the scarcity of CS data. Previous studies tend to rely on the model itself to implicitly learn semantic modeling during training, and resort to inefficient and costly manual annotations for these two challenges. To mitigate these limitations, we propose enhancing Large Language Models (LLMs) with a Mixture of Experts (MoE) speech projector, where each expert specializes in the semantic subspace of a specific language, enabling fine-grained modeling of speech features. Additionally, we introduce a multi-stage training paradigm that utilizes readily available monolingual automatic speech recognition (ASR) and monolingual ST data, facilitating speech-text alignment and improving translation capabilities. During training, we leverage a combination of language-specific loss and intra-group load balancing loss to guide the MoE speech projector in efficiently allocating tokens to the appropriate experts, across expert groups and within each group, respectively. To bridge the data gap across different training stages and improve adaptation to the CS scenario, we further employ a transition loss, enabling smooth transitions of data between stages, to effectively address the scarcity of high-quality CS speech translation data. Extensive experiments on widely used datasets demonstrate the effectiveness and generality of our approach.

arXiv:2511.10667v1 Announce Type: cross Abstract: Large language models (LLMs) often achieve impressive predictive accuracy, yet correctness alone does not imply genuine understanding. True LLM understanding, analogous to human expertise, requires making consistent, well-founded decisions across multiple instances and diverse domains, relying on relevant and domain-grounded decision factors. We introduce Structured Tabular Decision Simulations (STaDS), a suite of expert-like decision settings that evaluate LLMs as if they were professionals undertaking structured decision ``exams''. In this context, understanding is defined as the ability to identify and rely on the correct decision factors, features that determine outcomes within a domain. STaDS jointly assesses understanding through: (i) question and instruction comprehension, (ii) knowledge-based prediction, and (iii) reliance on relevant decision factors. By analyzing 9 frontier LLMs across 15 diverse decision settings, we find that (a) most models struggle to achieve consistently strong accuracy across diverse domains; (b) models can be accurate yet globally unfaithful, and there are frequent mismatches between stated rationales and factors driving predictions. Our findings highlight the need for global-level understanding evaluation protocols and advocate for novel frameworks that go beyond accuracy to enhance LLMs' understanding ability.

arXiv:2511.10664v1 Announce Type: cross Abstract: Large language models (LLMs) have achieved impressive results in high-resource languages like English, yet their effectiveness in low-resource and morphologically rich languages remains underexplored. In this paper, we present a comprehensive evaluation of seven cutting-edge LLMs -- including GPT-4o, GPT-4, Claude~3.5~Sonnet, LLaMA~3.1, Mistral~Large~2, LLaMA-2~Chat~13B, and Mistral~7B~Instruct -- on a new cross-lingual benchmark covering \textbf{Cantonese, Japanese, and Turkish}. Our benchmark spans four diverse tasks: open-domain question answering, document summarization, English-to-X translation, and culturally grounded dialogue. We combine \textbf{human evaluations} (rating fluency, factual accuracy, and cultural appropriateness) with automated metrics (e.g., BLEU, ROUGE) to assess model performance. Our results reveal that while the largest proprietary models (GPT-4o, GPT-4, Claude~3.5) generally lead across languages and tasks, significant gaps persist in culturally nuanced understanding and morphological generalization. Notably, GPT-4o demonstrates robust multilingual performance even on cross-lingual tasks, and Claude~3.5~Sonnet achieves competitive accuracy on knowledge and reasoning benchmarks. However, all models struggle to some extent with the unique linguistic challenges of each language, such as Turkish agglutinative morphology and Cantonese colloquialisms. Smaller open-source models (LLaMA-2~13B, Mistral~7B) lag substantially in fluency and accuracy, highlighting the resource disparity. We provide detailed quantitative results, qualitative error analysis, and discuss implications for developing more culturally aware and linguistically generalizable LLMs. Our benchmark and evaluation data are released to foster reproducibility and further research.