arXiv:2603.03911v1 Announce Type: new Abstract: Web security demands rapid response capabilities to evolving cyber threats. Agentic Artificial Intelligence (AI) promises automation, but the need for trustworthy security responses is of the utmost importance. This work investigates the role of semantic relations in extracting information for sensitive operational tasks, such as configuring security controls for mitigating threats. To this end, it proposes to leverage hypernym-hyponym textual relations to extract relevant information from Cyber Threat Intelligence (CTI) reports. By leveraging a neuro-symbolic approach, the multi-agent system automatically generates CLIPS code for an expert system creating firewall rules to block malicious network traffic. Experimental results show the superior performance of the hypernym-hyponym retrieval strategy compared to various baselines and the higher effectiveness of the agentic approach in mitigating threats.

arXiv:2603.04370v1 Announce Type: new Abstract: Conversational agents are increasingly deployed in knowledge-intensive settings, where correct behavior depends on retrieving and applying domain-specific knowledge from large, proprietary, and unstructured corpora during live interactions with users. Yet most existing benchmarks evaluate retrieval or tool use independently of each other, creating a gap in realistic, fully agentic evaluation over unstructured data in long-horizon interactions. We introduce $\tau$-Knowledge, an extension of $\tau$-Bench for evaluating agents in environments where success depends on coordinating external, natural-language knowledge with tool outputs to produce verifiable, policy-compliant state changes. Our new domain, $\tau$-Banking, models realistic fintech customer support workflows in which agents must navigate roughly 700 interconnected knowledge documents while executing tool-mediated account updates. Across embedding-based retrieval and terminal-based search, even frontier models with high reasoning budgets achieve only $\sim$25.5% pass^1, with reliability degrading sharply over repeated trials. Agents struggle to retrieve the correct documents from densely interlinked knowledge bases and to reason accurately over complex internal policies. Overall, $\tau$-Knowledge provides a realistic testbed for developing agents that integrate unstructured knowledge in human-facing deployments.

arXiv:2603.03295v1 Announce Type: cross Abstract: As large language models (LLMs) get integrated into human decision-making, they are increasingly choosing goals autonomously rather than only completing human-defined ones, assuming they will reflect human preferences. However, human-LLM similarity in goal selection remains largely untested. We directly assess the validity of LLMs as proxies for human goal selection in a controlled, open-ended learning task borrowed from cognitive science. Across four state-of-the-art models (GPT-5, Gemini 2.5 Pro, Claude Sonnet 4.5, and Centaur), we find substantial divergence from human behavior. While people gradually explore and learn to achieve goals with diversity across individuals, most models exploit a single identified solution (reward hacking) or show surprisingly low performance, with distinct patterns across models and little variability across instances of the same model. Even Centaur, explicitly trained to emulate humans in experimental settings, poorly captures people's goal selection. Chain-of-thought reasoning and persona steering provide limited improvements. These findings highlight the uniqueness of human goal selection, cautioning against replacing it with current models in applications such as personal assistance, scientific discovery, and policy research.

arXiv:2603.03301v1 Announce Type: cross Abstract: The rapid adoption of large language models (LLMs) has created demand for faster responses and lower costs. Semantic caching, reusing semantically similar requests via their embeddings, addresses this need but breaks classic cache assumptions and raises new challenges. In this paper, we explore offline policies for semantic caching, proving that implementing an optimal offline policy is NP-hard, and propose several polynomial-time heuristics. We also present online semantic aware cache policies that combine recency, frequency, and locality. Evaluations on diverse datasets show that while frequency based policies are strong baselines, our novel variant improves semantic accuracy. Our findings reveal effective strategies for current systems and highlight substantial headroom for future innovation. All code is open source.

arXiv:2603.03304v1 Announce Type: cross Abstract: We present a concise architecture for joint training on sentences and structured data while keeping knowledge and language representations separable. The model treats knowledge graphs and hypergraphs as structured instances with role slots and encodes them into a key-value repository that a language transformer can attend over. Attention is conditioned by journey-based role transport, which unifies edge-labeled KG traversal, hyperedge traversal, and sentence structure. We outline a dual-stream architecture, hierarchical layer groups with instance-local, neighborhood, and global mixing attention, retrieval over a separate repository, and multi-task objectives spanning masked language modeling, link prediction, and role-consistency denoising. The result is an explicit, inspectable separation between linguistic context and structured knowledge, while still enabling tight alignment through cross-attention.

arXiv:2603.03305v1 Announce Type: cross Abstract: Large language models (LLMs) are increasingly used to generate executable outputs, JSON objects, and API calls, where a single syntax error can make the output unusable. Constrained decoding enforces validity token-by-token via masking and renormalization, but it can distort generation when the model assigns low probability mass to valid continuations, pushing decoding toward locally valid yet semantically incorrect trajectories. We propose \emph{Draft-Conditioned Constrained Decoding (DCCD)}, a simple two-step, training-free inference procedure that decouples semantic planning from structural enforcement: an unconstrained draft is generated first, and constrained decoding is then applied, conditioned on this draft, to guarantee validity. We analyze DCCD through a KL-projection view, showing that draft conditioning increases feasible mass and reduces the cumulative "projection tax" induced by hard constraints, with an optional best-of-$K$ draft selection. Across structured reasoning benchmarks, DCCD improves strict structured accuracy by up to +24 percentage points over standard constrained decoding (e.g., 15.2\% to 39.0\% on GSM8K with a 1B model), and enables smaller model pairs to match or exceed much larger constrained baselines, yielding substantial gains in parameter efficiency.

arXiv:2603.03313v1 Announce Type: cross Abstract: Large Language Models (LLMs) exhibit a significant "embodiment gap", where their text-based representations fail to align with human sensorimotor experiences. This study systematically investigates whether and how task-specific fine-tuning can bridge this gap. Utilizing Representational Similarity Analysis (RSA) and dimension-specific correlation metrics, we demonstrate that the internal representations of LLMs can be steered toward more embodied, grounded patterns through fine-tuning. Furthermore, the results show that while sensorimotor improvements generalize robustly across languages and related sensory-motor dimensions, they are highly sensitive to the learning objective, failing to transfer across two disparate task formats.

arXiv:2603.03319v1 Announce Type: cross Abstract: Large Language Models (LLMs) are increasingly used as scalable evaluators of model outputs, but their preference judgments exhibit systematic biases and can diverge from human evaluations. Prior work on LLM-as-a-judge has largely focused on a small, predefined set of hypothesized biases, leaving open the problem of automatically discovering unknown drivers of LLM preferences. We address this gap by studying several embedding-level concept extraction methods for analyzing LLM judge behavior. We compare these methods in terms of interpretability and predictiveness, finding that sparse autoencoder-based approaches recover substantially more interpretable preference features than alternatives while remaining competitive in predicting LLM decisions. Using over 27k paired responses from multiple human preference datasets and judgments from three LLMs, we analyze LLM judgments and compare them to those of human annotators. Our method both validates existing results, such as the tendency for LLMs to prefer refusal of sensitive requests at higher rates than humans, and uncovers new trends across both general and domain-specific datasets, including biases toward responses that emphasize concreteness and empathy in approaching new situations, toward detail and formality in academic advice, and against legal guidance that promotes active steps like calling police and filing lawsuits. Our results show that automated concept discovery enables systematic analysis of LLM judge preferences without predefined bias taxonomies.

arXiv:2603.03320v1 Announce Type: cross Abstract: Effective communication often relies on aligning a message with an audience's narrative and worldview. Narrative shift involves transforming text to reflect a different narrative framework while preserving its original core message--a task we demonstrate is significantly challenging for current Large Language Models (LLMs). To address this, we propose a neurosymbolic approach grounded in social science theory and abductive reasoning. Our method automatically extracts rules to abduce the specific story elements needed to guide an LLM through a consistent and targeted narrative transformation. Across multiple LLMs, abduction-guided transformed stories shifted the narrative while maintaining the fidelity with the original story. For example, with GPT-4o we outperform the zero-shot LLM baseline by 55.88% for collectivistic to individualistic narrative shift while maintaining superior semantic similarity with the original stories (40.4% improvement in KL divergence). For individualistic to collectivistic transformation, we achieve comparable improvements. We show similar performance across both directions for Llama-4, and Grok-4 and competitive performance for Deepseek-R1.

arXiv:2603.03326v1 Announce Type: cross Abstract: Aligning Large Language Models (LLMs) with specific personas typically relies on expensive and monolithic Supervised Fine-Tuning (SFT) or RLHF. While effective, these methods require training distinct models for every target personality profile. Inference-time activation steering offers a parameter-efficient alternative, yet naive approaches fail to control multiple traits simultaneously due to destructive vector interference. In this work, we propose a modular framework for continuous, multi-dimensional personality control. Our key innovation is Sequential Adaptive Steering (SAS): a method that orthogonalizes steering vectors by training subsequent probes on the residual stream shifted by prior interventions. This approach transforms steering vectors into reusable primitives, allowing users to instantly synthesize complex, high-fidelity personality profiles by simply adjusting coefficients alpha. We validate our framework on the Big Five personality traits, demonstrating that it outperforms naive baselines in both goal adherence and coherence, enabling precise, holistic personality modulation without updating model parameters.

arXiv:2603.03332v1 Announce Type: cross Abstract: Chain-of-Thought (CoT) prompting has emerged as a foundational technique for eliciting reasoning from Large Language Models (LLMs), yet the robustness of this approach to corruptions in intermediate reasoning steps remains poorly understood. This paper presents a comprehensive empirical evaluation of LLM robustness to a structured taxonomy of 5 CoT perturbation types: \textit{MathError, UnitConversion, Sycophancy, SkippedSteps,} and \textit{ExtraSteps}. We evaluate 13 models spanning three orders of magnitude in parameter count (3B to 1.5T\footnote{Assumed parameter count of closed models}), testing their ability to complete mathematical reasoning tasks despite perturbations injected at different points in the reasoning chain. Our key findings reveal heterogeneous vulnerability patterns: MathError perturbations produce the most severe degradation in small models (50-60\% accuracy loss) but show strong scaling benefits; UnitConversion remains challenging across all scales (20-30\% loss even for largest models); ExtraSteps incur minimal accuracy degradation (0-6\%) regardless of scale; Sycophancy produces modest effects (7\% loss for small models); and SkippedSteps cause intermediate damage (15\% loss). Scaling relationships follow power-law patterns, with model size serving as a protective factor against some perturbations but offering limited defense against dimensional reasoning tasks. These findings have direct implications for deploying LLMs in multi-stage reasoning pipelines and underscore the necessity of task-specific robustness assessments and mitigation strategies. The code and results are available \href{https://github.com/Mystic-Slice/CoTPerturbation}{here}.

arXiv:2603.03464v1 Announce Type: cross Abstract: We introduce Graph Hopfield Networks, whose energy function couples associative memory retrieval with graph Laplacian smoothing for node classification. Gradient descent on this joint energy yields an iterative update interleaving Hopfield retrieval with Laplacian propagation. Memory retrieval provides regime-dependent benefits: up to 2.0~pp on sparse citation networks and up to 5 pp additional robustness under feature masking; the iterative energy-descent architecture itself is a strong inductive bias, with all variants (including the memory-disabled NoMem ablation) outperforming standard baselines on Amazon co-purchase graphs. Tuning enables graph sharpening for heterophilous benchmarks without architectural changes.

arXiv:2603.03485v1 Announce Type: cross Abstract: Recent video diffusion models have achieved impressive capabilities as large-scale generative world models. However, these models often struggle with fine-grained physical consistency, exhibiting physically implausible dynamics over time. In this work, we present \textbf{Phys4D}, a pipeline for learning physics-consistent 4D world representations from video diffusion models. Phys4D adopts \textbf{a three-stage training paradigm} that progressively lifts appearance-driven video diffusion models into physics-consistent 4D world representations. We first bootstrap robust geometry and motion representations through large-scale pseudo-supervised pretraining, establishing a foundation for 4D scene modeling. We then perform physics-grounded supervised fine-tuning using simulation-generated data, enforcing temporally consistent 4D dynamics. Finally, we apply simulation-grounded reinforcement learning to correct residual physical violations that are difficult to capture through explicit supervision. To evaluate fine-grained physical consistency beyond appearance-based metrics, we introduce a set of \textbf{4D world consistency evaluation} that probe geometric coherence, motion stability, and long-horizon physical plausibility. Experimental results demonstrate that Phys4D substantially improves fine-grained spatiotemporal and physical consistency compared to appearance-driven baselines, while maintaining strong generative performance. Our project page is available at https://sensational-brioche-7657e7.netlify.app/

arXiv:2603.03505v1 Announce Type: cross Abstract: State-of-the-art text-to-video (T2V) generators frequently violate physical laws despite high visual quality. We show this stems from insufficient physical constraints in prompts rather than model limitations: manually adding physics details reliably produces physically plausible videos, but requires expertise and does not scale. We present PhyPrompt, a two-stage reinforcement learning framework that automatically refines prompts for physically realistic generation. First, we fine-tune a large language model on a physics-focused Chain-of-Thought dataset to integrate principles like object motion and force interactions while preserving user intent. Second, we apply Group Relative Policy Optimization with a dynamic reward curriculum that initially prioritizes semantic fidelity, then progressively shifts toward physical commonsense. This curriculum achieves synergistic optimization: PhyPrompt-7B reaches 40.8\% joint success on VideoPhy2 (8.6pp gain), improving physical commonsense by 11pp (55.8\% to 66.8\%) while simultaneously increasing semantic adherence by 4.4pp (43.4\% to 47.8\%). Remarkably, our curriculum exceeds single-objective training on both metrics, demonstrating compositional prompt discovery beyond conventional multi-objective trade-offs. PhyPrompt outperforms GPT-4o (+3.8\% joint) and DeepSeek-V3 (+2.2\%, 100$\times$ larger) using only 7B parameters. The approach transfers zero-shot across diverse T2V architectures (Lavie, VideoCrafter2, CogVideoX-5B) with up to 16.8\% improvement, establishing that domain-specialized reinforcement learning with compositional curricula surpasses general-purpose scaling for physics-aware generation.

arXiv:2603.03507v1 Announce Type: cross Abstract: Adversarial attacks - input perturbations imperceptible to humans that fool neural networks - remain both a persistent failure mode in machine learning, and a phenomenon with mysterious origins. To shed light, we define and analyze a network's perceptual manifold (PM) for a class concept as the space of all inputs confidently assigned to that class by the network. We find, strikingly, that the dimensionalities of neural network PMs are orders of magnitude higher than those of natural human concepts. Since volume typically grows exponentially with dimension, this suggests exponential misalignment between machines and humans, with exponentially many inputs confidently assigned to concepts by machines but not humans. Furthermore, this provides a natural geometric hypothesis for the origin of adversarial examples: because a network's PM fills such a large region of input space, any input will be very close to any class concept's PM. Our hypothesis thus suggests that adversarial robustness cannot be attained without dimensional alignment of machine and human PMs, and therefore makes strong predictions: both robust accuracy and distance to any PM should be negatively correlated with the PM dimension. We confirmed these predictions across 18 different networks of varying robust accuracy. Crucially, we find even the most robust networks are still exponentially misaligned, and only the few PMs whose dimensionality approaches that of human concepts exhibit alignment to human perception. Our results connect the fields of alignment and adversarial examples, and suggest the curse of high dimensionality of machine PMs is a major impediment to adversarial robustness.

arXiv:2603.03517v1 Announce Type: cross Abstract: General-purpose large language models (LLMs) that rely on in-context learning do not reliably deliver the scientific understanding and performance required for drug discovery tasks. Simply increasing model size or introducing reasoning tokens does not yield significant performance gains. To address this gap, we introduce the MMAI Gym for Science, a one-stop shop molecular data formats and modalities as well as task-specific reasoning, training, and benchmarking recipes designed to teach foundation models the 'language of molecules' in order to solve practical drug discovery problems. We use MMAI Gym to train an efficient Liquid Foundation Model (LFM) for these applications, demonstrating that smaller, purpose-trained foundation models can outperform substantially larger general-purpose or specialist models on molecular benchmarks. Across essential drug discovery tasks - including molecular optimization, ADMET property prediction, retrosynthesis, drug-target activity prediction, and functional group reasoning - the resulting model achieves near specialist-level performance and, in the majority of settings, surpasses larger models, while remaining more efficient and broadly applicable in the domain.

arXiv:2603.03590v1 Announce Type: cross Abstract: In Multi-Agent Systems (MAS), agents are designed with social capabilities, allowing them to understand and reason about social concepts such as norms when interacting with others (e.g., inter-robot interactions). In Normative MAS (NorMAS), researchers study how norms develop, and how violations are detected and sanctioned. However, existing research in NorMAS use symbolic approaches (e.g., formal logic) for norm representation and reasoning whose application is limited to simplified environments. In contrast, Multimodal Large Language Models (MLLMs) present promising possibilities to develop software used by robots to identify and reason about norms in a wide variety of complex social situations embodied in text and images. However, prior work on norm reasoning have been limited to text-based scenarios. This paper investigates the norm reasoning competence of five MLLMs by evaluating their ability to answer norm-related questions based on thirty text-based and thirty image-based stories, and comparing their responses against humans. Our results show that MLLMs demonstrate superior performance in norm reasoning in text than in images. GPT-4o performs the best in both modalities offering the most promise for integration with MAS, followed by the free model Qwen-2.5VL. Additionally, all models find reasoning about complex norms challenging.

arXiv:2603.03684v1 Announce Type: cross Abstract: Recent developments show that AI can prove research-level theorems in mathematics, both formally and informally. This essay urges mathematicians to stay up-to-date with the technology, to consider the ways it will disrupt mathematical practice, and to respond appropriately to the challenges and opportunities we now face.

arXiv:2603.03704v1 Announce Type: cross Abstract: Robot planning in partially observable environments, where not all objects are known or visible, is a challenging problem, as it requires reasoning under uncertainty through partially observable Markov decision processes. During the execution of a computed plan, a robot may unexpectedly observe task-irrelevant objects, which are typically ignored by naive planners. In this work, we propose incorporating two types of common-sense knowledge: (1) certain objects are more likely to be found in specific locations; and (2) similar objects are likely to be co-located, while dissimilar objects are less likely to be found together. Manually engineering such knowledge is complex, so we explore leveraging the powerful common-sense reasoning capabilities of large language models (LLMs). Our planning and execution framework, CoCo-TAMP, introduces a hierarchical state estimation that uses LLM-guided information to shape the belief over task-relevant objects, enabling efficient solutions to long-horizon task and motion planning problems. In experiments, CoCo-TAMP achieves an average reduction of 62.7 in planning and execution time in simulation, and 72.6 in real-world demonstrations, compared to a baseline that does not incorporate either type of common-sense knowledge.

arXiv:2603.03727v1 Announce Type: cross Abstract: This paper studies how parents want to moderate children's interactions with Generative AI chatbots, with the goal of informing the design of future GenAI parental control tools. We first used an LLM to generate synthetic child-GenAI chatbot interaction scenarios and worked with four parents to validate their realism. From this dataset, we carefully selected 12 diverse examples that evoked varying levels of concern and were rated the most realistic. Each example included a prompt and a GenAI chatbot response. We presented these to parents (N=24) and asked whether they found them concerning, why, and how they would prefer the responses to be modified and communicated. Our findings reveal three key insights: (1) parents express concern about interactions that current GenAI chatbot parental controls neglect; (2) parents want fine-grained transparency and moderation at the conversation level; and (3) parents need personalized controls that adapt to their desired strategies and children's ages.