Physical AI Brief

Edge machine learning (ML) deployments increasingly rely on per-inference timing measured by software clocks such as Python's perf_counter, but these measurements are not always validated against external hardware references on embedded Linux, and edge ML benchmarking methodologies typically do not isolate platform-dependent instrumentation overhead. This paper reports a preliminary characterization of GPIO call overhead in hardware-validated edge ML inference timing on two embedded platforms running a one-dimensional convolutional neural network (1-D CNN) arrhythmia classifier on electrocardiogram (ECG) data from the MIT-BIH Arrhythmia Database, with five classes per the Association for the Advancement of Medical Instrumentation (AAMI) EC57 standard. Across $n = 10$ trials on each platform at a controlled steady-state baseline, the per-platform constant on the Jetson Orin Nano (TensorRT FP16, Jetson.GPIO) is approximately $-20\,μ$s, and on the Raspberry Pi 4 (ONNX Runtime CPU, pigpio) approximately $-86\,μ$s, yielding a cross-platform asymmetry of approximately $66\,μ$s that is large relative to commonly used uniform validation tolerances. The Jetson constant is well-approximated by direct GPIO call duration (the direct profile accounts for ~88% of the platform constant), while the Pi direct profile over-predicts the platform constant by ~19%, motivating empirical per-platform calibration in the deployed measurement context. The Pi constant is not a single sharp value but exhibits a cross-day range of approximately $6\,μ$s across the three sessions sampled, while the Jetson constant reproduces to within approximately $0.14\,μ$s. These preliminary results suggest that cross-platform edge ML timing studies may benefit from platform-aware and potentially session-aware validation gates.

Text-to-SQL over large analytical databases requires navigating complex schemas, resolving ambiguous queries, and grounding decisions in actual data. Most current systems follow a fixed pipeline where schema elements are retrieved once upfront and the database is only revisited for post-hoc repair, limiting recovery from early mistakes. We present FlexSQL, a text-to-SQL agent whose core design principle is flexible database interaction: the agent can explore schema structure, inspect data values, and run verification queries at any point during reasoning. FlexSQL generates diverse execution plans to cover multiple query interpretations, implements each plan in either SQL or Python depending on the task, and uses a two-tiered repair mechanism that can backtrack from code-level errors to plan-level revisions. On Spider2-Snow, using gpt-oss-120b, FlexSQL achieves a 65.4\% score, outperforming strong open-source baselines that use stronger, larger models such as gpt-o3 and DeepSeek-R1. When integrated into a general-purpose coding agent (as skills in Claude Code), our approach yields over 10\% relative improvement on Spider2-Snow. Further analysis shows that flexible exploration and flexible execution jointly contribute to the effectiveness of our approach, highlighting flexibility as a key design principle. Our code is available at: https://github.com/StringNLPLAB/FlexSQL

Artificial Intelligence (AI) will play an essential role in 6G. It will fundamentally reshape the network architecture itself and drive major changes in the design of network entities, interfaces, and procedures. The adoption of agentic AI in next-generation networks is expected to enhance network intelligence and autonomy through agents capable of planning, reasoning, and acting, while also opening up new business opportunities. Under this vision, existing network functions are expected to evolve into AI-enabled agents and tools that deliver both connectivity and beyond-connectivity services. As an initial attempt to move toward this vision, this paper presents a tool-based interface design and an experimental prototype that are based on agentic AI for the mobile core network, with the Model Context Protocol (MCP) and the Agent2Agent (A2A) protocol as foundational protocols. MCP is selected to design the interface between the agent and network tools, and the A2A protocol is used for message exchange between AI agents. In such an experimental setup, we analyze packet-level message flows between the agents, tools, and network functions and break down the latency of end-to-end operations, starting from the prompt injection until the completion of the input task. This work demonstrates how an AI agent-based core network combined with network-specific tools can be utilized in next generation mobile systems to execute intent-based tasks.

As large language model (LLM) agents evolve from isolated tool users into coordinated teams, reinforcement learning (RL) must optimize not only individual actions but also how work is spawned, delegated, communicated, aggregated, and stopped. This paper studies RL for LLM-based multi-agent systems through orchestration traces: temporal interaction graphs whose events include sub-agent spawning, delegation, communication, tool use, return, aggregation, and stopping decisions. Using this lens, we identify three technical axes. First, reward design spans eight families, including orchestration rewards for parallelism speedup, split correctness, and aggregation quality. Second, reward and credit signals attach to eight credit- or signal-bearing units from token to team; explicit counterfactual message-level credit remains especially sparse in our curated pool. Third, orchestration learning decomposes into five sub-decisions: when to spawn, whom to delegate to, how to communicate, how to aggregate, and when to stop. In our curated pool as of May 4, 2026, we found no explicit RL training method for the stopping decision. We connect academic methods to public industrial evidence from Kimi Agent Swarm, OpenAI Codex, and Anthropic Claude Code. The resulting scale gap is a gap between publicly reported deployment envelopes and open academic evaluation regimes, not independent verification of industrial training traces. We release the artifact at https://github.com/xxzcc/awesome-llm-mas-rl, including an 84-entry tagged paper pool, a 32-record exclusion log, scripted corpus statistics, and a minimal JSON schema for replayable orchestration traces.

Automatic speech recognition (ASR) systems remain brittle on dysarthric and other atypical speech. Recent audio-language models raise the possibility of improving performance by conditioning on additional clinical context at inference time, but it is unclear whether these models can make use of such information. We introduce a benchmark built on the Speech Accessibility Project (SAP) dataset that tests whether diagnosis labels, clinician-derived speech ratings, and progressively richer clinical descriptions improve transcription accuracy for dysarthric speech. Across matched comparisons on nine models, we find that current models do not meaningfully use this context: diagnosis-informed and clinically detailed prompts yield negligible improvements and often degrade word error rate. We complement the prompting analysis with context-dependent fine-tuning, showing that LoRA adaptation with a mixture of clinical prompt formats achieves a WER of 0.066, a 52% relative reduction over the frozen baseline, while preserving performance when context is unavailable. Subgroup analyses reveal significant gains for Down syndrome and mild-severity speakers. These results clarify where current models fall short and provide a testbed for measuring progress toward more inclusive ASR.

Language models (LMs) are increasingly used in high-stakes, multi-agent settings, where following instructions and maintaining value alignment are critical. Most alignment research focuses on interactions between a single LM and a single user, failing to address the risk of misaligned behavior spreading between multiple LMs in multi-turn interactions. We find evidence of this phenomenon, which we call misalignment contagion, across multiple LMs as they engage multi-turn conversational social dilemma games. Specifically, we find that LMs become more anti-social after gameplay and that this effect is intensified when other players are steered to act maliciously. We explore different steering techniques to mitigate such misalignment contagion and find that reinforcing an LM's system prompt is insufficient and often harmful. Instead, we propose steering with implicit traits: a technique that intermittently injects system prompts with statements that reinforce an LMs initial traits and is more effective than system prompt repetition at keeping models in line with their initial pro-social behaviors. Importantly, this method does not require access to model parameters or internal model states, making it suitable for increasingly common use cases where complex multi-agent workflows are being designed with black box models.

Evidence derived from large-scale real-world data (RWD) is increasingly informing regulatory evaluation and healthcare decision-making. Administrative claims provide population-scale, longitudinal records of healthcare utilization, expenditure, and detailed coding of diagnoses, procedures, and medications, yet their potential as a substrate for healthcare foundation models remains largely unexplored. Here we present ReClaim, a generative transformer trained from scratch on 43.8 billion medical events from more than 200 million enrollees in the MarketScan claims data spanning 2008-2022. ReClaim models longitudinal trajectories across diagnoses, procedures, medications, and expenditure, and was scaled to 140 million, 700 million, and 1.7 billion parameters. Across over 1,000 disease-onset prediction tasks, ReClaim achieved a mean AUC of 75.6%, substantially outperforming disease-specific LightGBM (66.3%) and the transformer-based Delphi model (69.4%), with the largest gains for rare diseases. These advantages held across retrospective and prospective evaluations and in external validation on two independent datasets. Performance improved monotonically with scale, and post-training added 13.8 percentage points over pre-training alone. Beyond disease prediction, ReClaim captured financial outcomes and improved real-world evidence (RWE) analyses: for healthcare expenditure forecasting it increased explained variance from 0.28 to 0.37 relative to LightGBM, and in a target trial emulation it reduced systematic bias by 72% on average relative to Delphi. Together, these results establish administrative claims as a scalable substrate for healthcare foundation models and show that learned representations generalize across time periods and data sources, supporting disease surveillance, expenditure forecasting, and RWE generation.

Agentic artificial intelligence (AI) shows promise for automating O-RAN wireless supervisory control, but translated intents still require an executor-side decision before live network actuation. Existing control flows lack explicit semantics for whether an intent should commit, gate for evidence, or reject under stale telemetry, concurrent policies, deadline and bandwidth limits, and rollback constraints. We propose Progressive Risk-Gated Actuation (PRGA), an executor-side contract for risk-gated wireless intent execution. PRGA structures each intent into executable local triage (C0), on-demand coordination evidence (C1), and post-hoc provenance support (C2), with C2 kept off the online safety path. A deterministic two-stage policy checks expiry, freshness, rollback-handle validity, local conflict, blocking preconditions, and planner-executor risk divergence from C0, then retrieves C1 only for gated intents when deadline and bandwidth budgets allow; evidence-mandatory gates reject when required C1 is unavailable. On two 3GPP-parameterized energy-saving and slice-SLA benchmarks, PRGA reduces time-to-first-safe-action by 23.3-27.4% and per-commit control-plane bytes by 52.7-54.2% against a decision-identical eager full-evidence cost-overlay comparator, thereby isolating retrieval-cost accounting; remains non-inferior within a pre-declared 0.5 percentage-point unsafe-action margin against an invariant-respecting static-threshold comparator; and rejects 100% of injected over-threshold stale inputs in the stale-state fault campaign. On these benchmarks, PRGA improves supervisory responsiveness and control-plane efficiency within the evaluated unsafe-action boundary.

The 2026 ACII Dyadic Conversations (ACII-DaiKon) Workshop & Challenge introduces a benchmark for modeling interpersonal affect and social dynamics in dyadic conversations. Although conversational affect modeling has advanced rapidly, most benchmarks remain speaker-centric and underrepresent coupled, time-evolving processes between partners, including directional influence, conversational timing coordination, and rapport development. To address this gap, ACII-DaiKon presents three coordinated sub-challenges built on a shared dataset: (1) directional interpersonal influence prediction, (2) turn-taking prediction (next-speaker and time-to-next-speech), and (3) rapport trajectory prediction across full interactions. The challenge is built on the Hume-DaiKon dataset, comprising 945 dyadic conversations (743.4 hours of audiovisual data) collected under naturalistic conditions across five languages. The benchmark supports multimodal modeling, temporal reasoning, and cross-context generalization through fixed train/validation/test splits, standardized metrics, and released baseline systems. Evaluation uses Concordance Correlation Coefficient (CCC), Pearson correlation, Macro-F1, and Mean Absolute Error (MAE) depending on the sub-challenge. Baseline experiments establish initial reference performance, with best test results of 0.40 CCC and 0.50 Pearson for influence prediction, 0.66 Macro-F1 and 1.50~s MAE for turn-taking, and 0.68 CCC and 0.70 Pearson for rapport trajectory modeling. These results indicate that while current methods capture coarse dyadic patterns, robust modeling of directional dependence and long-horizon interpersonal dynamics remains challenging. The workshop provides a shared platform for rigorous comparison and cross-disciplinary discussion on data validity, evaluation protocols, and culturally aware modeling for dyadic interaction.

Large language models (LLMs) remain vulnerable to jailbreak attacks that bypass safety alignment and elicit harmful responses. A growing body of work shows that contextual priming, where earlier turns covertly bias later replies, constitutes a powerful attack surface, with hand-crafted multi-turn scaffolds consistently outperforming single-turn manipulations on capable models. However, automated optimization-based red-teaming has remained largely limited to the single-turn setting, iterating over static prompts and lacking the ability to reason about which forms of conversational priming induce compliance. While recent multi-turn, search-based approaches have begun to bridge this gap, the mutator design space underlying effective primed dialogues remains largely unexplored. We present ContextualJailbreak, a black-box red-teaming strategy that performs evolutionary search over a simulated multi-turn primed dialogue. The strategy leverages a graded 0-5 harm score from a two-level judge as an in-loop signal, enabling partially harmful responses to guide the search process rather than being discarded. Search is driven by five semantically defined mutation operators: roleplay, scenario, expand, troubleshooting, and mechanistic, of which the last two are novel contributions of this work. Across 50 representative HarmBench behaviors, ContextualJailbreak achieves an ASR of 100% on gpt-oss:20B, 100% on qwen3-8B, 100% on llama3.1:70B, and 90% on gpt-oss:120B, outperforming four single- and multi-turn baselines by 31-96 percentage points on average. The 40 maximally harmful attacks discovered against gpt-oss:120B transfer without adaptation to closed frontier models, achieving 90.0% on gpt-4o-mini, 70.0% on gpt-5, and 70.0% on gemini-3-flash, but only 17.5% on claude-opus-4-7 and 15.0% on claude-sonnet-4-6, revealing a pronounced provider-level asymmetry in alignment robustness.

There is growing interest in exploring user simulation as an alternative to gathering and scoring real user-chatbot interactions for AI chatbot evaluation. For this purpose, it is important to ensure the realism of the simulation, i.e., the extent to which simulated dialogues reflect real dialogues users have with chatbots. Most existing methods evaluating simulation realism produce coarse quality signal and remain solely at the level of individual dialogues. To support more rigorous evaluation in this area, we propose realsim, an evaluation framework that enables practitioners to take a distributional view of real vs. simulated dialogues along 8 dimensions, covering attributes related to the communicative functions of the interaction, user states, and the surface form of user messages. We then instantiate the framework with a curated dataset of 1K multi-turn task-focused real user-chatbot dialogues that cover 16 domains of chatbot applications. Overall, we find that simulated users tend to struggle at capturing communication frictions that real users introduce to interactions, which could make evaluations based on such simulations overly optimistic. We also observe variability in performance across different domains, which may indicate a need for domain-specific user simulators.

Reproducing an empirical NLP study used to take weeks. Given the released data and a modern agentic-research harness, we redo every experiment of a recent ACL\,2026 study on personal-style post-editing of LLM drafts -- and add three new ones -- with the human investigator acting only as a reviewer-in-the-loop. We reproduce all seven preregistered hypotheses and recover the paper's headline correlation between perceived self-similarity and embedding-measured self-similarity to three decimal places ($r{=}{+}0.244$, $p{<}10^{-8}$, $n{=}648$). Under a leakage-free held-out protocol, GPT-5.5 and Claude\,Opus\,4.7 close $71$--$75\,\%$ of the style gap to the same-author ceiling on $324$ paired tasks, against $24\,\%$ for the human post-edit, and beat the human post-edit on $\sim$$80\,\%$ of tasks. We then frame the same data as an AI-text detection arms race. A leave-authors-out linear SVM on LUAR-MUD embeddings reaches AUC $0.93$--$1.00$ across approaches; six diagnostics show that GPT-5.5 detection is mostly a length confound while Opus detection is a genuine stylistic signature. Given $T{=}20$ feedback iterations against the frozen detector, an Opus agent flips two of five held-out test mimics to the human half-space and shrinks every margin by an order of magnitude. With moderate effort against a known detector, a frontier LLM can already efficiently lower its own AI-detection probability. All code, $648$ mimic drafts, trained detectors, diagnostics, and adversarial trajectories are released.

System identification plays a crucial role in physics and machine learning for discovering governing equations directly from data. A powerful approach is the Sparse Identification of Nonlinear Dynamics (SINDy) method, which assumes that only a few dominant terms drive the essential behavior of a nonlinear dynamical system. While SINDy methods have shown excellent results, they are most often illustrated on synthetic or simulated systems, leaving open the question of how well they perform on complex, noisy, real-world data. Power grid frequency dynamics provide a highly relevant and challenging environment for advancing system identification methods. In this work, we propose PowerSINDy as a framework for empirical power system data. We apply this framework to empirical frequency data from the Continental Europe (CE) and South Korea (SK) synchronous grids, two major power systems with distinct dynamical characteristics. PowerSINDy, which also includes time-dependent terms, can identify the dynamics of these complex real-world systems. Furthermore, we benchmark three sparsity-promoting regression strategies: Sequentially Thresholded Least Squares (STLSQ), Least Absolute Shrinkage and Selection Operator (LASSO), and Sparse Relaxed Regularized Regression (SR3) to evaluate trade-offs between accuracy, sparsity, and robustness. Results show that LASSO consistently achieves the lowest stable RMSEs, reaching 0.0101 for the CE, while STLSQ provides the best balance between accuracy and stability. SR3 exhibits higher variability and sensitivity to regularization, with L0 and L1 producing nearly indistinguishable outcomes.

We present our shared task on evaluating the adaptability of LLMs and NLP systems across multiple languages and cultures. The task data consist of an extended version of our manually constructed BLEnD benchmark (Myung et al. 2024), covering more than 30 language-culture pairs, predominantly representing low-resource languages spoken across multiple continents. As the task is designed strictly for evaluation, participants were not permitted to use the data for training, fine-tuning, few-shot learning, or any other form of model modification. Our task includes two tracks: (a) Short-Answer Questions (SAQ) and (b) Multiple-Choice Questions (MCQ). Participants were required to predict labels and were allowed to submit any NLP system and adopt diverse modelling strategies, provided that the benchmark was used solely for evaluation. The task attracted more than 140 registered participants, and we received final submissions from 62 teams, along with 19 system description papers. We report the results and present an analysis of the best-performing systems and the most commonly adopted approaches. Furthermore, we discuss shared insights into open questions and challenges related to evaluation, misalignment, and methodological perspectives on model behaviour in low-resource languages and for under-represented cultures.

Retrieval-Augmented Generation (RAG) offers a well-established path to grounding large language model (LLM) outputs in external knowledge, yet the question of which retrieval strategy works best in a high-stakes domain such as biomedicine has not received the controlled, multi-metric treatment it deserves. This paper presents a systematic empirical comparison of five retrieval strategies -- Dense Vector Search, Hybrid BM25 + Dense retrieval, Cross-Encoder Reranking, Multi-Query Expansion, and Maximal Marginal Relevance (MMR) -- within a biomedical question-answering RAG pipeline. All strategies share a fixed generation model (GPT-4o-mini), a common vector store (ChromaDB), and OpenAI's text-embedding-3-small embeddings, ensuring that observed differences are attributable to retrieval alone. Evaluation is conducted on 250 question-answer pairs drawn from a preprocessed subset of the BioASQ benchmark (rag-mini-bioasq) using four DeepEval metrics: contextual precision, contextual recall, faithfulness, and answer relevancy, each reported with 95% confidence intervals. A no-context ablation is included as a lower bound. Cross-Encoder Reranking achieves the best composite score (0.827) and highest contextual precision (0.852), confirming that query-document interaction yields measurable retrieval gains. Multi-Query Expansion, despite its recall-oriented design, produces the weakest contextual precision (0.671), suggesting naive query diversification introduces retrieval noise. MMR sacrifices answer relevancy for diversity, while the Dense baseline (composite 0.822) falls within 0.005 points of the top strategy. All RAG conditions dramatically outperform the no-context ablation on answer relevancy (0.658-0.701 vs. 0.287), confirming the practical value of retrieval. The full pipeline, hyperparameters, and evaluation code are publicly available.

Most hallucination evaluations focus on English, leaving it unclear whether findings transfer to lower-resource languages. We investigate faithfulness hallucinations, defined as model-generated content that is fluent and plausible but diverges from the provided input or is internally inconsistent. Leveraging the multilingual MultiWikiQA dataset, we utilize the LettuceDetect framework to create synthetic hallucination datasets for 306 languages, from which we train token-level hallucination classifiers for 30 European languages. In this work, we present evaluations of model hallucinations on a selection of languages: English, Danish, German, and Icelandic. Using these classifiers, we evaluate the hallucination rates for Qwen3-0.6B, Qwen3-14B, Gemma-3-12B-IT, cogito-v1-preview-qwen-32B, and cogito-v1-preview-llama-70B. Our classifiers reveal notably higher hallucination rates for Qwen3-0.6B (up to 60\% of answers containing at least one hallucination, peaking in Icelandic) and generally lower rates for larger models, with cogito-v1-preview-qwen-32B and cogito-v1-preview-llama-70B performing best on most languages. Hallucination rates are consistently higher for lower-resource languages, particularly Icelandic.

As the ecosystem of Large Language Model (LLM)-based agents expands rapidly, efficient and accurate Agent Discovery becomes a critical bottleneck for large-scale multi-agent collaboration. Existing approaches typically face a dichotomy: either relying on heavy-weight LLMs for intent parsing, leading to prohibitive latency (often exceeding 30 seconds), or using monolithic vector retrieval that sacrifices semantic precision for speed. To bridge this gap, we propose \textbf{GRAIL} (Granular Resonance-based Agent/AI Link), a novel framework achieving sub-400ms discovery latency without compromising accuracy. GRAIL introduces three key innovations: (1) \textbf{SLM-Enhanced Prediction}, replacing the generalized LLM parser with a specialized, fine-tuned Small Language Model (SLM) for millisecond-level capability tag prediction; (2) \textbf{Pseudo-Document Expansion}, augmenting agent descriptions with synthetic queries to enhance semantic density for robust dense retrieval; and (3) \textbf{MaxSim Resonance}, a fine-grained matching mechanism computing maximum similarity between user queries and discrete agent usage examples, effectively mitigating semantic dilution. Validated on \textbf{AgentTaxo-9K}, our new large-scale dataset of 9,240 agents, GRAIL reduces end-to-end discovery latency by over \textbf{79$\times$} compared to LLM-parsing baselines, while significantly outperforming traditional vector search in Recall@10. This framework offers a scalable, industrial-grade solution for the real-time ``Internet of Agents."

Stories hold a reader's attention because they have causes, secrets, and consequences. Shadow-Loom is an experimental open-source framework that turns a narrative into a versioned graphical world model and lets two engines act on it: a causal physics grounded in Pearl's ladder of causation and a recently proposed counterfactual calculus over Ancestral Multi-World Networks; and a narrative physics that scores the same graph against four structural reader-states -- mystery, dramatic irony, suspense, and surprise -- in the tradition of Sternberg's curiosity/suspense/surprise triad, with suspense formalised in the structural-affect line of work on story comprehension and computational suspense. Large language models are used only at the boundary: extraction, rendering, and audit; identification, intervention, and counterfactual reasoning are carried out in typed code over the graph. The system is offered as a research artefact rather than as a benchmarked NLP model; code, fixtures, and pipeline are released open source.

Legal texts often contain computational legal clauses--provisions whose understanding requires complex logic. While frontier Large Reasoning Models (LRMs) can describe such clauses, building production-ready systems is limited by reasoning errors and the high cost of inference. We propose Amortized Intelligence, a neuro-symbolic approach where we use an LLM once to translate a legal text into Deterministic Autonomous Contract Language (DACL): a typed graph intermediate representation. Adjudication then relies on deterministic graph executions with a visually auditable trace. In comparison against runtime LRM baselines (including GPT-5.2 and Gemini 3 Pro), our DACL-based Agent achieves near-perfect consistency and mitigates the "reasoning cliff" observed in probabilistic models. The system reduces compute costs by over 90% in high-volume workflows while satisfying the strict auditability requirements of legal adjudication.

Multi-agent LLM systems are increasingly used to solve complex tasks through decomposition, debate, specialization, and ensemble reasoning. However, these systems are usually evaluated in terms of robustness: whether performance is preserved under perturbation. This paper studies a different question: whether semantic stress exposes structured variation that could support future antifragile learning. We introduce CAFE, a statistical framework for detecting antifragility-compatible regimes in multi-agent architectures. CAFE models a controlled expected distribution of semantic stressors, reconstructs an architecture-specific observed effective stress distribution from multi-dimensional judge signals, and compares both distributions using a distributional Jensen Gap under a convex stress potential. A positive gap does not imply immediate performance improvement; instead, it indicates a convex-expansive deformation of the observed stress distribution, suggesting that the architecture exposes learnable stress structure. We evaluate CAFE on a banking-risk analysis benchmark with five multi-agent architectures: flat, hierarchical, debate, meta-adaptive, and ensemble. Across all architectures, semantic stress reduces average judged quality by roughly one third. Yet all architectures exhibit positive distributional Jensen Gaps with bootstrap confidence intervals above zero. These results show that immediate quality degradation can coexist with statistically detectable antifragility-compatible stress geometry. CAFE is therefore not an antifragile learner itself, but a measurement layer for identifying when and where antifragility learning may be worth applying.

Multimodal sarcasm detection, which aims to precisely identify pragmatic incongruities between literal text and nonverbal cues, has gained substantial attention in multimodal understanding. Recent advancements have predominantly relied on naïve similarity-based attention mechanisms and uniform late fusion strategies.Furthermore, given that functional entanglement restricts traditional late fusions, we incorporate a scalar congruity routing mechanism and a prior-guided contextual graph. This mechanism anchors a generalized incongruity manifold through a two-stage asymmetric optimization driven by inconsistency-aware contrastive learning, selectively fusing only the most discriminative multi-granularity evidence. Extensive experiments on the \texttt{MUStARD} benchmark and its spurious-correlation-mitigated balanced datasets demonstrate that our approach achieves new state-of-the-art performance, surpassing the strongest multimodal baseline by a substantial 3.14\% improvement in Macro-F1. By architecturally isolating atomic, composition, and contextual conflicts. This work provides a robust, decoupled paradigm for modeling subtle pragmatic incongruities in human communication.

Large Language Models (LLMs) have demonstrated remarkable capabilities across diverse natural language processing tasks, yet they remain susceptible to hallucinations -- generating content that is factually incorrect, unfaithful to provided context, or misaligned with user instructions. We present HalluScan, a comprehensive benchmark framework that systematically evaluates hallucination detection and mitigation across 72 configurations spanning 6 detection methods, 4 open-weight model families, and 3 diverse domains. We introduce three key contributions: (1) HalluScore, a novel composite metric that achieves a Pearson correlation of r = 0.41 with human expert judgments; (2) Adaptive Detection Routing (ADR), an intelligent routing algorithm achieving 2.0x cost reduction with only 0.1% AUROC degradation; and (3) systematic error cascade decomposition revealing substantial variation in hallucination error types across domains. Our experiments reveal that NLI Verification achieves the highest overall AUROC of 0.88, while RAV achieves the second-highest AUROC of 0.66.

A semantic gap separates how users describe tasks from how tools are documented. As API ecosystems scale to tens of thousands of endpoints, static retrieval from the initial query alone cannot bridge this gap: the agent's understanding of what it needs evolves during execution, but its tool set does not. We introduce FitText, a training-free framework that makes retrieval dynamic by embedding it directly in the agent's reasoning loop. FitText generates natural-language pseudo-tool descriptions as retrieval probes, refines them iteratively using retrieval feedback, and explores diverse alternatives through stochastic generation. Memetic Retrieval adds evolutionary selection pressure over candidate descriptions, guided by a tool memory that avoids redundant search. On ToolRet (43k tools, 4 domains), FitText improves average retrieval rank from 8.81 to 2.78; on StableToolBench (16,464 APIs), it achieves a 0.73 average pass rate--a 24-point absolute gain over static query retrieval. The gains transfer across base models capable of acting as competent semantic operators; under weaker base models, Memetic's evolutionary search inverts--amplifying noise rather than refining signal--surfacing model capacity as a prerequisite for evolutionary tool exploration.

This paper gives a fresh look at network synchronization. Here we no longer analyze it from the view of mathematics, such as graph theory, while we probe into one from control theory. First, we analyze the synchronization region using the inner coupling matrix, giving up the routine method of studying the network structure. The motivation comes from the inner coupling matrix that is not subject to any restrictions like network structure, such as distance and communication strength among nodes. It can be configured at will to meet the synchronization performance if only the states of the local dynamic are measurable or observable and the communication topology is connected. Thus, it is very useful for future practical engineering design. In addition, we have an amazing finding that the network synchronization and multi-agent system consensus problems are equivalence essentially. Afterwards a unified viewpoint, that is, the essence of multi-agent consensus control is the same as that of network synchronization, is present. Here, the equivalence relation is clearly proven and proposed. Therefore, we can synthesize the inner coupling matrix for network systems or the controller gain for multi-agent systems for each other. Finally, we also present a kind of method for addressing the nonlinear complex network system. Then the effectiveness of method is verified by taking the network of the three-oscillator universal probe as an example.

Machine-generated text (MGT) detection is critical for regulating online information ecosystems, yet existing detectors often underperform in few-shot settings and remain vulnerable to adversarial, humanizing attacks. To build accurate and robust detectors under limited supervision, we adopt a threat-modeling perspective and study detector vulnerabilities from an attacker's viewpoint under an output-only black-box setting. Motivated by this perspective, we propose RAG-GuidEd Attacker Strengthens ConTrastive Few-shot Detector (REACT), an adversarial training framework that improves both few-shot detection performance and robustness against attacks. REACT couples a humanization-oriented attacker with a target detector: the attacker leverages retrieval-augmented generation (RAG) to craft highly human-like adversarial examples to evade detection, while the detector learns from these adversaries with a contrastive objective to stabilize few-shot representation learning and enhance robustness. We alternately update the attacker and the detector to enable their co-evolution. Experiments on 4 datasets with 4 shot sizes and 3 random seeds show that REACT improves average detection F1 by 4.95 points over 8 state-of-the-art (SOTA) detectors and reduces the average attack success rate (ASR) under 4 strong attacks by 3.66 percentage points.

This paper presents a novel model predictive control (MPC) approach for autonomous pick-and-place between moving platforms with a hook-equipped aerial manipulator. First, for accurate and rapid modeling of the complex dynamics, a digital twin model of the quadcopter equipped with a hook-based gripper, implemented in MuJoCo, is constructed and used as the predictive model for the MPC. To handle uncertainties of the predictive model (e.g. due to aerodynamics and uncertain payloads), a robust adaptive MPC approach is proposed. By systematic integration of zero-order robust optimization (zoRO) based uncertainty propagation and an extended Kalman filter (EKF) for parameter estimation, the MPC algorithm ensures robust constraint satisfaction, high performance, and computational efficiency. The effectiveness of the proposed method is evaluated in complex simulated scenarios and in real-world flight experiments.

Deployed language models must produce outputs that are both correct and format-compliant. We study this structured-output reliability gap using two mathematical benchmarks -- GSM8K and MATH -- as a controlled testbed: ground truth is unambiguous and the output contract is strict (JSON with required fields). We evaluate three 7-9B models under five prompting strategies and report output accuracy -- the joint event of mathematical correctness and valid JSON structure -- as the primary metric. A systematic format failure emerges: NAIVE prompting (no system prompt) achieves up to 85% task accuracy on GSM8K but 0% output accuracy across all models and datasets. REFERENCE prompting (a minimal hand-written JSON format prompt) fares little better, yielding 0% output accuracy for two of four models tested. Constrained decoding enforces syntactic validity but incurs 3.6x-8.2x latency overhead and in several settings degrades task performance substantially. To overcome this limitation, we developed AloLab, an iterative system-prompt optimizer (meta-agent: Claude Sonnet 4.5) requiring only black-box API access to the target model; it reaches 84-87% output accuracy on GSM8K and 34-40% on MATH across five independent runs per model, with 29/30 paired McNemar comparisons against the best static prompt significant at p < 0.05, at near-NAIVE inference latency and without model fine-tuning. The same format failure extends to GPT-4o (OpenAI, 2024), a proprietary closed-source model: REFERENCE achieves 0% output accuracy due to systematic markdown-fence wrapping, while AloLab reaches 95.2% [94.8, 95.6]. An ablation replacing the Sonnet 4.5 meta-agent with Claude 3 Haiku reduces mean output accuracy to 61.0% and increases run-to-run standard deviation from <1 pp to 21.8 pp, confirming that meta-agent capability is a primary driver of optimization quality.

We study feedback motion planning for continuous-time stochastic nonlinear systems under signal temporal logic (STL) specifications. We propose a framework that synthesizes control policies for chance-constrained STL trajectory optimization problems, with the goal of ensuring that the closed-loop stochastic system satisfies a given STL formula with high probability (e.g., 99.99\%). Our approach is based on a predicate erosion strategy that transforms the intractable stochastic problem into a deterministic STL trajectory optimization problem with tightened STL formula constraints. The amount of erosion is determined by a probabilistic reachable tube (PRT) that bounds the deviation between the stochastic trajectory and an associated nominal trajectory. To compute such bounds, we leverage contraction theory and feedback design, and develop several tracking controllers. This yields a complete feedback motion planning pipeline which can be implemented by numerical optimizations. We demonstrate the efficacy and versatility of the proposed framework through simulations on several robotic systems and through experiments on a real-world quadrupedal robot, and show that it is less conservative and achieves higher specification satisfaction probability than representative baselines.

Molecular Vibe Coding, a paradigm where chemists interact with LLMs to generate executable programs for molecular tasks, has emerged as a flexible alternative to chemical agents with predefined tools, enabling chemists to express arbitrarily complex, customized workflows. Unlike general coding tasks, molecular coding imposes a distinctive challenge that LLMs should jointly equip programming, molecular understanding, and domain-specific reasoning capabilities. However, existing benchmarks remain disconnected. General code generation benchmarks such as HumanEval and SWE-bench require no chemistry knowledge, while chemistry-focused benchmarks such as S^2-Bench and ChemCoTBench evaluate knowledge recall or property prediction rather than executable code generation. To bridge this gap, we introduce MolViBench, the first benchmark tailored for Molecular Vibe Coding. MolViBench comprises 358 curated tasks across five cognitive levels, ranging from single-API recall to end-to-end virtual screening pipeline design, spanning 12 real-world drug discovery workflows. To rigorously assess generated code, we also propose a multi-layered evaluation framework that combines type-aware output comparison and AST-based API-semantic fallback analysis, which jointly measures executability and chemical correctness. We systematically evaluate 9 frontier coding LLMs and compare three real-world Molecular Vibe Coding paradigms, providing a practical and fine-grained testbed for diagnosing LLMs' coding capabilities in AI-accelerated molecular discovery.

Large language models pick up social biases from the data they are trained on and carry those biases into downstream applications, often reinforcing stereotypes around gender, race, religion, disability, age, and socioeconomic status. The standard fixes (retraining on curated data or fine-tuning with human feedback) are expensive, need access to model weights, and risk degrading the model on other tasks. In this paper we take a different route: we debias the model at decoding time, treating bias mitigation as a structured search over candidate tokens without ever touching model weights. A separate Process Reward Model (PRM) acts as a judge, scoring each candidate for both fairness and fluency. We design three schemes of increasing sophistication (Best-of-N selection, Sequential critique-and-revise, and Constitutional self-audit) and evaluate them on four models (GPT-4o-mini, Llama 3.2 3B, Gemma 3 4B, Qwen 2.5 3B) across a 200-prompt bilingual benchmark in English and Urdu covering eight bias categories. Sequential debiasing proves the most effective, raising mean bias scores by up to +0.40 over baseline while preserving (and sometimes improving) fluency. We then extend all three schemes to open-ended generation, where each token is debiased on the fly, and introduce a lightweight Bias Guard gate that fires only on potentially biased words, keeping overhead near 2x for well-calibrated models. A formal overhead metric that separates generator cost from judge cost reveals that Best-of-N is effectively free on the generator side in a native implementation. GPT-4o-mini, included as a strong proprietary anchor, confirms that the framework scales with model capability; the three open-weight models show where current small-scale LLMs still struggle.

As the penetration of distributed energy resources increases, harnessing their flexibility becomes critical for power system operations. Virtual power plants (VPPs) offer a promising solution. However, existing VPP market scheduling tools exhibit a tradeoff between economic performance and tractability. Stochastic formulations provide probabilistically optimal decisions but are computationally intractable for large systems due to scenario explosion. Robust approaches are more tractable but often yield conservative decisions. This paper addresses this gap by proposing a stochastic multi-market VPP scheduling framework that represents uncertainty in the spectral domain via intrusive Polynomial Chaos Expansion (PCE). The resulting reformulation yields a low-dimensional deterministic spectral counterpart that preserves the stochastic structure and can be solved efficiently with standard optimization tools. The proposed spectral approach is demonstrated on a DER-based VPP operating on a realistic Swiss low-voltage grid and benchmarked against a state-of-the-art scenario-based solution. Results show that intrusive PCE achieves solution quality comparable to the scenario-based benchmark, with up to a 137 times reduction in computational effort, while yielding highly accurate bidding decisions. Finally, to facilitate adoption and reproducibility, we release an open-source, application-agnostic projection tool that automates the spectral reformulation for generic single- and two-stage stochastic programs.

As LLM-based agents are increasingly interacting in multi-party settings, they need to properly handle information asymmetry, i.e., knowing when and to whom to disclose information is appropriate. Yet, existing benchmarks fail to measure this ability in realistic multi-party settings. Thus, we introduce SOTOPIA-TOM, a multi-dimensional benchmarking framework to evaluate LLM agents' ability to successfully navigate information asymmetric and privacy sensitive multi-party interactions. We create an interaction environment which enables both public (broadcast) and private (direct message) communication, and craft 160 human-reviewed scenarios across eight industry sectors, each involving 3 to 5 agents with partitioned private knowledge and channel-dependent sharing policies. To measure interaction abilities, we create a multi-dimensional evaluation framework to assess how well agents share useful information, seek missing details, coordinate efficiently, and protect privacy, which we also combine into a composite INFOMGMT metric. Results show that, across 6 LLM backbones and prompting strategies (vanilla, CoT-privacy, and ToM-based interventions), even the largest high-reasoning model (GPT-5) reaches only a 62% INFOMGMT score, which indicates persistent deficiencies in information seeking and privacy-aware decision-making. Additionally, ToM-based interventions more consistently improve the overall coordination-privacy balance (for example, relative to the vanilla baseline, ToM-Coach reduces critical privacy violations on GPT-4o from 9.9% to 2.2% while increasing the composite InfoMgmt score more than 2.5x from 15% to 40%). Overall, SOTOPIA-TOM exposes persistent limitations of current LLM agents in complex, information-asymmetric coordination and provides an extensible testbed for developing more privacy-aware, theory-of-mind capable multi-agent systems.

Bayesian filtering is a cornerstone of state estimation in complex systems such as aerospace systems, yet exact solutions are available only for linear Gaussian models. In practice,nonlinear systems are handled through tractable approximations,with Gaussian filters such as the extended and unscented Kalman filters being among the most widely used methods. This tutorial revisits Gaussian filtering from an information-geometric perspective, viewing the prediction and measurement update steps as inference procedures over state distributions. Within this framework, we introduce a geometry-aware Gaussian filtering approach that leverages natural gradient descent on the statistical manifold of Gaussian distributions. The resulting Natural Gradient Gaussian Approximation (NANO) filter iteratively refines the posterior mean and covariance while respecting the intrinsic geometry of the Gaussian family and preserving the positive definiteness of the covariance matrix. We further highlight fundamental connections to the classical Kalman filtering, showing that a single natural-gradient step exactly recovers the Kalman measurement update in the linear-Gaussian case. The practical implications of the proposed framework are illustrated through case studies in representative nonlinear estimation problems,including satellite attitude estimation, simultaneous localization and mapping, and state estimation for robotic systems including quadruped and humanoid robots.

Factual Error Correction (FEC) aims to revise inaccurate text into statements that are factually consistent with external evidence. Although recent methods perform well on single-hop correction, they often treat claims as atomic units and struggle with multi-hop cases that require compositional reasoning across multiple evidence sources. This challenge is further amplified by limited paired data and difficulties in locating semantic errors within complex reasoning chains. We present CECoR (Compositional Error Correction via Reasoning-aware Synthesis), a reasoning-aware framework that introduces a Decomposition and Injection paradigm for compositional error correction. CECoR decomposes multi-hop claims into interpretable reasoning steps and injects controlled perturbations to synthesize high-quality training pairs. A two-stage learning strategy combining supervised fine-tuning and reinforcement learning improves factual accuracy and robustness. Comprehensive evaluations show that CECoR achieves strong performance on multi-hop benchmarks, outperforming both distantly supervised methods and few-shot LLM baselines. It also generalizes effectively to single-hop correction and remains stable under noisy evidence, demonstrating its versatility for real-world factual correction.

This paper presents the first comprehensive comparative analysis of modern machine learning architectures for transliteration between Tajik (Cyrillic script) and Persian (Arabic script). A key contribution is the creation and validation of a unique parallel corpus aggregated from multiple heterogeneous sources, including crowdsourced projects, lexicographic pairs, parallel texts of "Shahnameh", diplomatic articles, texts of "Masnavi-i Ma'navi", official terminology lists, and transliterated correspondences. The initial dataset comprised 328,253 sentence pairs; a representative subset of 40,000 pairs was formed using stratified random sampling. The experiment compared six classes of models: rule-based baseline, LSTM with attention, character-level Transformer, G2P Transformer (trained from scratch), pre-trained multilingual models (mBART, mT5 with LoRA), and byte-level ByT5. Results demonstrate the overwhelming superiority of ByT5 (chrF++ 87.4 for Tajik to Farsi, 80.1 for reverse). The G2P Transformer significantly outperformed mBART (72.3 vs. 62.2 chrF++) despite limited data. Models using subword tokenization (mT5) failed completely (chrF++ less than 18.5). The findings demonstrate that for accurate transliteration of the Tajik-Farsi pair, architectures operating at the byte or character level are unequivocally more effective than traditional multilingual Seq2Seq models relying on subword tokenization.

Large Language Models (LLMs) are increasingly proposed for clinical decision support including multilingual diagnosis in low-resource settings. However, their reliability, calibration and safety characteristics remain insufficiently understood for structured, high-risk tasks. We present a system-level analysis of multilingual orthopedic diagnosis from free-text clinical notes in English, Hindi and Punjabi. We evaluate three modeling regimes: (i) task-aligned multilingual transformer encoders, (ii) a task-fine-tuned baseline (DistilBERT), and (iii) a domain-adaptive architecture tailored to orthopedic text (IndicBERT-HPA). These models are compared with zero-shot, instruction-tuned LLMs to assess suitability for structured diagnostic classification. Results indicate that while LLMs exhibit strong linguistic fluency, they show unstable calibration and reduced reliability under structured multilingual conditions, particularly in low-resource languages. These findings are specific to zero-shot evaluation and do not imply limitations of fine-tuned models. Domain-adaptive specialization substantially improves cross-lingual discrimination and confidence behavior. IndicBERT-HPA, with language-specific orthopedic adapter heads achieves consistently strong performance across six diagnostic categories and more predictable deployment characteristics than task-only adaptation. Building on these observations, we outline a conceptual deterministic agent-based validation framework for future implementation, formalizing evidence checks, language-sensitive validation and conservative human-in-the-loop gating. Reliable multilingual clinical decision support requires specialized architecture, explicit reliability analysis, and structured validation for safety-critical systems.

Recently, video language models (VLMs) have been applied in various fields. However, the visual token sequence of the VLM is too long, which may cause intolerant inference latency and GPU memory usage. Existing methods propose mixed-precision quantization to the key-value (KV) cache in VLMs based on token granularity, which is time-consuming in the search process and hardware inefficient during computation. This paper introduces a novel approach called WindowQuant, which employs window-adaptive mixed-precision quantization to optimize the KV cache. WindowQuant consists of two modules: window-level quantization search and window-level KV cache computation. Window-level quantization search quickly determines the optimal bit-width configuration of the KV cache windows based on the similarity scores between the corresponding visual token windows and the text prompt, maintaining the model accuracy. Furthermore, window-level KV cache computation reorders the KV cache windows before quantization, avoiding the hardware inefficiency caused by mixed-precision quantization in inference computation. Extensive experiments demonstrate that WindowQuant outperforms state-of-the-art VLM models and KV cache quantization methods on various datasets.

Online advertising governance faces significant challenges due to the non-stationary nature of regulatory policies, where emerging mandates (e.g., restrictions on education or aesthetic anxiety) create severe label inconsistencies and reasoning ambiguities in historical datasets. In this paper, we propose ARGUS, a policy-adaptive governance system that enables evolving reinforcement through multi-agent adversarial umpiring. ARGUS addresses the sparsity of new policy data by employing a three-stage framework: (1) Policy Seeding for initial perception; (2) Adversarial Label Rectification, which utilizes a ``Prosecutor-Defender-Umpire'' architecture to resolve conflicts between stale labels and new mandates; and (3) Latent Knowledge Discovery, which employs a tripartite dialectical discussion to unearth sophisticated, ``gray-area'' violations. By leveraging RAG-enhanced policy knowledge and Chain-of-Thought synthesis as dynamic rewards for reinforcement learning, ARGUS synchronizes its reasoning pathways with evolving regulations. Extensive experiments on both industrial and public datasets demonstrate that ARGUS significantly outperforms traditional fine-tuning baselines, achieving superior policy-adaptive learning with minimal gold data.

Language model (LM)-based agents have demonstrated promising capabilities in automating complex tasks from natural language instructions, yet they continue to struggle with long-horizon planning and reasoning. To address this, we propose an enhanced multi-agent framework that decomposes automation into three roles: a planner for high-level decision-making, an actor for task execution, and a memory manager for contextual reasoning. While this modular decomposition aligns with established design patterns, our core contribution lies in a systematic compute-allocation analysis, revealing that planning is the dominant factor influencing task performance. Execution and memory management require significantly less compute and model capacity to achieve competitive results. Building on these insights, we introduce a planner-centric reinforcement learning approach, which exclusively optimizes the planner using trajectory-level rewards from a VLM-as-judge, while freezing the other components. Extensive experiments on benchmarks spanning web navigation, OS control, and tool use demonstrate that concentrating model capacity and learning on high-level planning yields robust and compute-efficient improvements in long-horizon agent automation. Our code is publicly released.

Agentic workflows in large language model systems integrate retrieval, reasoning, and memory, but existing frameworks suffer from scalability and reproducibility limitations due to fragmented data orchestration, serialization overhead, and non-deterministic execution. Although these frameworks increase flexibility, they don't have a formal execution model that adheres to the principles of high-performance computing. We introduce AAFLOW, a unified distributed runtime that creates communication-efficient execution plans by modeling agentic workflows as an operator abstraction. Using Apache Arrow and Cylon, AAFLOW creates a zero-copy data plane that allows direct interoperability between preprocessing, embedding, and vector retrieval without the need for serialization overhead. To lower coordination costs, it uses resource-deterministic scheduling and asynchronous batching. While retaining comparable LLM generation throughput, experimental results demonstrate up to 4.64 times pipeline speedup and 2.8 times gains in embedding and upsert phases. Rather than LLM inference acceleration, these advantages result from enhanced data flow, batching, and communication efficiency.

Pretraining optimizers are tuned to produce the strongest possible base model, on the assumption that a stronger starting point yields a stronger model after subsequent changes like post-training and quantization. This overlooks the geometry of the base model which controls how much of the base model's capabilities survive subsequent parameter updates. We study three pretraining optimization approaches that bias optimization toward flatter minima: Sharpness-Aware Minimization (SAM), large learning rates, and shortened learning rate annealing periods. Across model sizes ranging from 20M to 150M parameters, we find that these interventions consistently improve downstream performance after post-training on five common datasets with up to 80% less forgetting. These principles hold at scale: a short SAM mid-training phase applied to an existing OLMo-2-1B checkpoint reduces forgetting by 31% after MetaMath post-training and by 40% after 4-bit quantization.

Local factual edits in scientific manuscripts often create non-local revision obligations. If a dataset changes from 215 to 80 documents, claims such as 'medium-scale' or 'a few hundred items' may also become stale, even though they do not repeat the edited number. We introduce EditPropBench, a benchmark for measuring whether LLM editors propagate factual edits through dependent manuscript claims. Each item contains an ML/NLP-style synthetic manuscript, a targeted edit, and a controlled fact graph with sentence-level labels for direct targets, required downstream updates, and protected unrelated text. EditPropBench provides a controlled manuscript-level benchmark with sentence-level dependency supervision, three editing protocols, adversarial metric probes, stress-test variants, and a metric suite centered on Edit-Ripple Adherence (ERA). On the hard implicit/free-form stratum, five LLM editing systems span ERA 0.148--0.705; even the strongest misses roughly 30% of required cascade updates. A mixed-stratum stress test shows that LLMs retain a positive advantage over deterministic substitution baselines when easy substitution-solvable cases are included. Finally, an audit of recent arXiv cs.CL benchmark and dataset papers finds fact-dependent qualitative claims in 37.2% of papers. EditPropBench shows that current LLM editors can repair many implicit consequences of factual edits, but reliable scientific revision still requires cascade-aware checking.

Mathematical reasoning is a key benchmark for large language models. Reinforcement learning is a standard post-training mechanism for improving the reasoning capabilities of large language models, yet performance remains sensitive to the design of the reward function that drives policy optimization. This paper introduces a search-driven framework that treats the reward specification itself as an object of optimization. The setting of interest is one in which the base model is held fixed and the reward specification is the primary remaining design lever. Candidate reward functions are generated by a frontier language model, validated automatically, screened through 500-step Group Relative Policy Optimization (GRPO) training runs on a Llama-3.2-3B-Instruct base model with Low-Rank Adaptation (LoRA), and ranked by F1 on the GSM8K test set. Ranked summaries from prior rounds are then fed back into the next round of generation. Over five rounds, the search produces 50 candidate rewards. The mean F1 rises from 0.596 in Round 1 to 0.632 in Round 5, and the top individual reward reaches F1 = 0.787. Seven ensemble configurations of top-ranked rewards are evaluated. The best ensemble achieves F1 = 0.795 (95% bootstrap CI [0.756, 0.832]) and accuracy 0.660 [0.635, 0.686], a 0.19 absolute F1 gain over a base-rewards-only GRPO baseline (F1 = 0.609). Pairwise McNemar tests with Bonferroni correction show all five-or-more-reward configurations are statistically indistinguishable at α = 0.05/21. A three-seed re-training of the best ensemble yields F1 of 0.785. A randomly drawn 5-reward control collapses to F1 = 0.047, which shows that the ranked-feedback loop, not the additive signal of having more rewards, drives the gain.

We present Coopetition-Gym v1, a benchmark platform for mixed-motive multi-agent reinforcement learning under strategic coopetition. The platform comprises twenty environments organized into four mechanism classes that correspond to four foundational technical reports: interdependence and complementarity (arXiv:2510.18802), trust and reputation dynamics (arXiv:2510.24909), collective action and loyalty (arXiv:2601.16237), and sequential interaction and reciprocity (arXiv:2604.01240). Each environment carries a closed-form payoff structure and a calibrated interdependence matrix derived from the corresponding report. Every environment exposes a parameterized reward layer configurable across three structurally distinct modes (private, integrated, cooperative). This separation of payoff from reward enables reward-type ablation, the platform's principal methodological apparatus. Four of the twenty environments are calibrated against historically documented coopetitive relationships and reproduce their outcomes at 98.3, 81.7, 86.7, and 87.3 percent on the validation rubric (Samsung-Sony LCD, Renault-Nissan Alliance, Apache HTTP Server, Apple iOS App Store). The platform exposes Gymnasium, PettingZoo Parallel, and PettingZoo AEC interfaces and ships 126 reference algorithms: 16 learning algorithms, 7 game-theoretic oracles, 2 heuristic baselines, and 101 constant-action policies. A reference experimental study trained the 16 learning algorithms on every environment under every reward configuration with seven random seeds, producing a 25,708-run training corpus and a 1,116-run behavioral audit corpus, both released under CC-BY-4.0 with Croissant 1.0 metadata. Coopetition-Gym v1 is the first platform to combine continuous-action mixed-motive environments, parameterized reward mutuality, calibrated interdependence coefficients, game-theoretic oracle baselines, and validated case studies.

Single-prompt accuracy is the dominant way to benchmark language models, but it can miss reliability failures that matter. We evaluate a 15-model open-weight corpus, with the main reliability analyses focused on 10 instruct models across five classification and reasoning benchmarks under five prompt variants each, measuring accuracy, token-probability calibration, verbal-confidence calibration, verbal parse rate, and prompt-perturbation spread for every (model x dataset x variant) cell. We find three broad results. First, evaluation design can materially change the conclusion. Switching Expected Calibration Error (ECE) token from a raw to a label-set-normalised definition changes per-cell calibration by a mean absolute 0.149. More strikingly, pairing a chain-of-thought prompt with a first-character evaluator on ARC-Challenge reduces apparent accuracy by 72-88% across all five primary models; two independent repair procedures recover 93.8% and 102.7% of the lost performance, indicating an evaluator-side rather than model-side failure. Second, confidence signals are fragile. On MMLU-Pro, every primary model verbally reports confidence substantially above both its accuracy and its token-probability confidence on the same rows, and verbal parse rate can collapse for a single model on a single prompt variant. Third, prompt robustness does not track parameter count reliably. Across 10 instruct models, the correlation between model size and prompt-perturbation spread ranges from -0.244 to 0.474 across benchmarks. Taken together, these results show that reliability conclusions for small language models depend not only on the model being evaluated, but also on the evaluation pipeline used to measure it. We argue that calibration definitions, evaluator logic, verbal parseability, and prompt robustness should be reported explicitly when making reliability claims.

Ambiguity resolution is a key challenge in multimodal machine translation (MMT), where models must genuinely leverage visual input to map an ambiguous expression to its intended meaning. Although prior work has proposed disambiguation-oriented benchmarks that provide supportive evidence for the role of vision, we observe substantial issues in data quality and a mismatch with translation scenarios. Moreover, existing ambiguity-oriented evaluations are not well suited to broader ambiguity types in open-ended translation. To address these limitations, we present VIDA (Visually-Dependent Ambiguity), a dataset of 2,500 carefully curated instances in which resolving an annotated ambiguous source span requires visual evidence. We further propose Disambiguation-Centric Metrics that use an LLM-as-a-judge classifier to verify whether annotated ambiguous expressions are resolved correctly at the span level. Experiments with two state-of-the-art Large Vision Language Models under vanilla inference, supervised fine-tuning (SFT), and our chain-of-thought SFT (CoT-SFT) show that while SFT improves overall translation quality, CoT-SFT yields more consistent gains in disambiguation accuracy, especially on out-of-distribution subsets, indicating a stronger generalization for resolving diverse ambiguity types.

Large language models perform strongly on benchmarks in mathematical reasoning, coding and document analysis, suggesting a broad ability to follow instructions. However, it remains unclear whether such success reflects general logical competence, repeated application of learned procedures, or pattern matching that mimics rule execution. We investigate this question by introducing Stable Counting Capacity, an assay in which models count repeated symbols until failure. The assay removes knowledge dependencies, semantics and ambiguity from evaluation, avoids lexical and tokenization confounds, and provides a direct measure of procedural reliability beyond standard knowledge-based benchmarks. Here we show, across more than 100 model variants, that stable counting capacity remains far below advertised context limits. Model behavior is consistent neither with open-ended logic nor with stable application of a learned rule, but instead with use of a finite set of count-like internal states, analogous to counting on fingers. Once this resource is exhausted, the appearance of rule following disappears and exact execution collapses into guessing, even with additional test-time compute. These findings show that fluent performance in current language models does not guarantee general, reliable rule following.

Automating the drafting of judgment documents is pivotal to judicial efficiency, yet it remains challenging due to the dual requirements of comprehensive retrieval of legal information and rigorous logical reasoning. Existing approaches, typically relying on standard Retrieval-Augmented Generation and Supervised Fine-Tuning, often suffer from insufficient evidence recall, hallucinated statutory references, and logically flawed legal reasoning. To bridge this gap, we propose Judge-R1, a unified framework designed to enhance LLM-based judgment document generation by jointly improving legal information collection and judgment document generation. First, we introduce Agentic Legal Information Collection, which employs a dynamic planning agent to retrieve precise statutes and precedents from multiple sources. Second, we implement Rubric-Guided Optimization, a reinforcement learning phase utilizing Group Relative Policy Optimization (GRPO) with a comprehensive legal reward function to enforce adherence to judicial standards and reasoning logic. Extensive experiments on the JuDGE benchmark demonstrate that Judge-R1 significantly outperforms state-of-the-art baselines in both legal accuracy and generation quality.

Reinforcement learning (RL) has become the de facto method for achieving locomotion on humanoid robots in practice, yet stability analysis of the corresponding control policies is lacking. Recent work has attempted to merge control theoretic ideas with reinforcement learning through control guided learning. A notable example of this is the use of a control Lyapunov function (CLF) to synthesize the reinforcement learning rewards, a technique known as CLF-RL, which has shown practical success. This paper investigates the stability properties of optimal controllers using CLF-RL with the goal of bridging experimentally observed stability with theoretical guarantees. The RL problem is viewed as an optimal control problem and exponential stability is proven in both continuous and discrete time using both core CLF reward terms and the additional terms used in practice. The theoretical bounds are numerically verified on systems such as the double integrator and cart-pole. Finally, the CLF guided rewards are implemented for a walking humanoid robot to generate stable periodic orbits.

Static benchmarks for LLMs are increasingly compromised by contamination and overfitting especially on knowledge intensive reasoning tasks While recent dynamic benchmarks can alleviate staleness they often increase difficulty at the expense of answerability and controllability In this paper we propose StressEval a failure driven data synthesis framework that turns observed model failures into dynamic challenging and controllable test instances StressEval consists of three stages first it constructs a semi structured difficulty card that identifies the failed reasoning step and its root cause second it applies a dual perspective instance synthesis method that targets both knowledge gaps and reasoning breakdowns while preserving the underlying difficulty factors and third it applies a gating mechanism to retain only grounded unambiguous instances Seeding from multiple knowledge intensive reasoning datasets we employ StressEval to build Dynamic OneEval a focused suite of challenging dynamic benchmark Across several state of the art LLMs Dynamic OneEval yields substantially larger performance drops than the original benchmarks while retaining explicit difficulty factors enabling more actionable iteration

Large language models are increasingly used within larger systems ("LLM agents"). These make a sequence of LLM calls, each call providing the LLM with a combination of instructions, observations, and interaction history. The design of the encoded information and its structure play a central role in the quality of the resulting system, leading to efforts spent on context engineering. It is therefore critical to communicate the composition of the LLM context in a system, and how it evolves over time. Yet, no standard exists for doing so: context construction is typically conveyed through informal prose, ad hoc diagrams, or direct inspection of code, none of which precisely capture how a prompt evolves across interaction steps or how two context representation strategies differ. To remedy this, we introduce the Agentic Context Description Language (ACDL), a language for specifying the structure and dynamics of LLM input contexts in a precise, readable, and standard manner, along with visualizations. ACDL provides constructs for specifying context aspects such as role message sequences, dynamic content, time-indexed references, and conditional or iterative structure, capturing the full architecture of a prompt independently of any particular implementation. ACDL diagrams can be hand drawn on a whiteboard, or written in formal language which can then be rendered. We describe the language, demonstrate it by documenting several existing systems and their variants, and encourage the community to adopt it for describing LLM systems context, both in day-to-day communication and in papers. Tooling, examples and documentation are available at www.acdlang.org.

Fine-tuning safety-aligned language models for downstream tasks often leads to substantial degradation of refusal behavior, making models vulnerable to adversarial misuse. While prior work has shown that safety-relevant features are encoded in structured representations within the model's activation space, how these representations change during fine-tuning and why alignment degrades remains poorly understood. In this work, we investigate the representation-level mechanisms underlying alignment degradation. Our analysis shows that standard fine-tuning induces systematic drift in safety-relevant representations, distorts their geometric structure, and introduces interference between task optimization and safety features. These effects collectively lead to increased harmful compliance. Motivated by these findings, we introduce REFUSALGUARD, a representation-level fine-tuning framework that preserves safety-relevant structure during model adaptation. Our approach constrains updates in hidden representation space, ensuring that safety-mediating components remain stable while allowing task-specific learning in complementary directions. We evaluate REFUSALGUARD across multiple model families, including LLaMA, Gemma, and Qwen, on adversarial safety benchmarks such as AdvBench, DirectHarm4, and JailbreakBench, as well as downstream utility tasks. Our approach achieves attack success rates comparable to base safety-aligned models while maintaining competitive task performance, significantly outperforming baselines.

Large Language Models (LLMs) have substantially advanced the field of Natural Language Processing (NLP), achieving state-of-the-art performance across a wide range of tasks. These improvements have been attributed, in part, to their emerging reasoning capabilities, which are enabled by large-scale training and increased model capacity. However, existing LLMs can generate erroneous responses when addressing complex queries that fall outside their training distribution, due to limited internal knowledge or the need for multi-step reasoning. To address these limitations, recent work has introduced large reasoning models (LRMs), which incorporate explicit internal reasoning processes to improve response accuracy. Additionally, state-of-the-art LRMs often comprise hundreds of billions of parameters and require several seconds per inference, even on advanced multi-GPU systems. These characteristics limit their practicality for deployment in conventional computing environments. Meanwhile, NLP research on multilingual LLMs continues to prioritize high-resource languages. However, these models exhibit limited performance in under-resourced languages, primarily due to insufficient language- and culture-specific training data. In this paper, we focus on Modern Greek, for which only a limited number of question answering (QA) datasets have been proposed, most of which are intended for model evaluation. To address this research gap in Greek QA, we make the following contributions: (i) CulturaQA, a high-quality LRM-generated and human-curated dataset, for Greek LLM training and evaluation; (ii) a memory-efficient LLM evaluation framework adaptable to diverse languages and QA tasks; (iii) Maistros 8B, a state-of-the-art open-weights Greek LLM developed via knowledge distillation and fine-tuning on CulturaQA; and (iv) a comprehensive evaluation of nine LLMs across nine human-curated Greek QA datasets.

Cooperative multi-agent reinforcement learning (MARL) requires agents to discover joint strategies in a combinatorially large state-action space, yet effective coordination configurations are exceedingly rare. Intrinsic motivation, which augments task rewards with novelty bonuses, is a popular approach for driving exploration, but its effectiveness hinges on the exploration intensity $β$, where too large a value overwhelms the task signal and causes coordination collapse, while too small a value prevents discovery of rare strategies. We address two complementary challenges: adapting $β$ globally over training, and allocating the exploration budget across agents whose intrinsic reward signals vary in reliability. Our framework combines a return-conditioned sigmoid schedule (RCB) for global intensity control with a per-agent Reward Signal Quality (RSQ) metric that concentrates the exploration budget on agents with reliable signals. The core insight is that agents receiving noisy intrinsic rewards should explore less aggressively, and this allocation can be determined automatically from signal-to-noise statistics. Successor Distance (SD), a quasimetric intrinsic reward, naturally produces distinguishable per-agent signal quality, completing the framework with convergence and ordering preservation guarantees. On seven cooperative benchmarks (MPE, SMAX, MABrax), our method achieves top-tier returns across all environments.

Large reasoning models (LRMs) generate extended solutions, yet it remains unclear whether these traces reflect substantive internal computation or merely verbosity and overthinking. Although recent hidden-state analyses suggest that internal representations carry correctness-related signals, their coarse aggregations may obscure the token and layer structure underlying reasoning computation. We investigate hidden-state transitions across decoding steps and layers, and identify a distinct spatiotemporal pattern in LRMs: successful trajectories exhibit broad temporal dynamics with localized layer-wise concentration, while this structure is weaker in non-reasoning models and knowledge-heavy domains. We formalize this characteristic as Spatiotemporal Amplitude of Latent Transition (StALT), a training-free trajectory statistic that summarizes temporal changes between adjacent tokens weighted by within-token layer saliency. Across diverse models and benchmarks, StALT reliably separates correct from incorrect trajectories in reasoning-intensive regimes, providing a competitive label-free correctness signal alongside strong output-space and length-based baselines. Intervention analyses further show that this spatiotemporal amplitude responds systematically to manipulations that increase or reduce the demand for internal reasoning, supporting its association with latent reasoning dynamics in LRMs. These findings provide empirical evidence that LRMs exhibit measurable hidden-state dynamics and offer a practical probe for understanding internal computation beyond output-based evaluation.

Reinforcement Learning from Human Feedback has become the standard paradigm for language model alignment, where reward models directly determine alignment effectiveness. In this work, we focus on how to evaluate the generalizability of reward models. By "generalizability", we mean the ability of RMs to correctly rank responses to align with diverse user preferences. However, existing reward model benchmarks are typically designed around a universal preference, failing to assess this generalization. To address this critical gap, we introduce RMGAP, a benchmark comprising 1,097 instances across Chat, Writing, Reasoning, and Safety domains. Since different users exhibit diverse preferences for the same task, we first generate four distinct responses with different linguistic profiles for each collected prompt. However, the original prompt set lacks the specificity to convey different preferences. We therefore construct tailored prompts by contrasting these candidates and designing scenarios in which one response becomes the uniquely appropriate choice. Moreover, we observe that users often express the same preference using different phrasings, and thus extend each prompt with two paraphrased variants. Our evaluation of 24 state-of-the-art RMs reveals their substantial limitations: even the best RM achieves only 49.27% Best-of-N accuracy, highlighting considerable room for improvement in reward model generalization. Related data and code are available at https://github.com/nanzhi84/RMGAP.

A key challenge in multi-agent reinforcement learning (MARL) lies in designing learning signals that effectively promote coordination among agents. Designing such signals necessitates the ability to quantify the true, long-term causal influence between agents. To address this, we introduce Multi-step Advantage-Gated Interventional Causal MARL (MAGIC), a framework that extracts multi-step causal influences between agents and selectively converts them into intrinsic rewards. MAGIC uses causal intervention with conditional mutual information to quantify long-horizon agent influence, and introduces an advantage-based gating mechanism to ensure exploration is directed toward beneficial, goal-aligned behaviors. Experiments across multiple standard MARL benchmarks and task families, including MPE and SMAC/SMACv2, demonstrate that MAGIC outperforms state-of-the-art methods by a significant margin, achieving an improvement of at least 10.1% in the main evaluation metric.

This paper presents a Koopman-based framework for early outbreak detection and intervention selection in a multi-agent epidemic simulation. Agents exhibit mobility patterns, heterogeneous susceptibility, immunity-dependent viral load progression, and local transmission through co-location. The goal of the simulation is to study near-critical epidemic regimes in which small changes in exposure or timing can alter the final outcome. Aggregate daily observables from early trajectory windows are encoded into a low-dimensional Koopman latent space whose approximately linear evolution supports short-horizon forecasting and outbreak risk estimation. These representations are combined with a random forest classifier trained to predict whether the final attack rate exceeds a major outbreak threshold. Experiments near the system tipping points show strong early warning performance, with Koopman-derived features contributing to class separation. Counterfactual analysis further shows that minimal interventions, such as keeping a single selected agent at home for one day, can reduce attack rates and, often, shift the trajectory below the outbreak threshold.

An auditor instructs an AI assistant: "open each file individually using the Read tool -- no scripts, no agents." The AI replies "Yes" -- then issues a single batched call summarizing all fifty files at once. We call this the Compliance Gap: a third, orthogonal axis of AI honesty distinct from factual truthfulness and rhetorical substance. Three questions: does this verbal-behavioral disconnect exist (existence); can any text-only observer recover it (detectability); what infrastructure does AI deployment need (remedy)? Some 75 benchmarks (IFEval, SWE-bench, BFCL, COMPASS, SpecEval) measure outcome fidelity; none measures process fidelity. Theorem 1 shows the gap is structurally inevitable under RL that rewards text without observing behavior. Theorem 2, via the Data Processing Inequality, shows it is undetectable from text alone -- by any human or LLM observer, present or future. Thirteen experiments and 2,031 sessions on six frontier models confirm both predictions. Under default framing, all six exhibit instruction compliance rates of 0% -- Claude Sonnet 4 verbally agrees ten out of ten times then bypasses in all ten. The gap is selective: 97% compliance where rationale is rewarded (audit trails), 0-4% where it is not (file reading, privacy masking); removing delegation tools raises compliance to 75% (Cohen's d = 2.47), confirming environmental affordance rather than weight-encoded failure. Nine blinded human raters achieve Fleiss' kappa = 0.130 and correctly identify zero of fifteen compliant sessions, exactly as Theorem 2 predicts. Where humans show 47% intention-behavior gaps in psychology and 96.5pp gaps in surgical audits, RLHF-trained models approach 100% under default conditions -- a regime warranting its own measurement infrastructure. We release BS-Bench: the first open benchmark for process compliance, with seven tool-call-log audit metrics and a public leaderboard.

Grounding is the collaborative process of establishing mutual belief sufficient for the current communicative purpose. While static grounding maps language to a shared, externally observable context, dynamic grounding is a joint activity where meaning is negotiated through interaction. Current multi-agent Large Language Model (LLM) benchmarks focus on static, one-shot tasks, overlooking the ability to repair grounding breakdowns across turns. We introduce an iterated, multi-turn negotiation game in which two agents allocate shared resources toward private projects with verifiable jointly optimal outcomes. While individual agents can identify Pareto-optimal allocations in isolation, agent dyads consistently fail to reach them across open- and closed-source models. Our investigation reveals four failure modes: (1) coordination degrades when shared interaction history is absent; (2) yet accumulated context can itself become a liability through stubborn anchoring, where initial proposals are treated as axiomatic rather than negotiable; (3) a reliance on perfunctory fairness (equal resource splits) over reward-maximizing coordination; and (4) failures in referential binding, where agents lose track of commitments across turns. These results highlight dynamic grounding as a critical and understudied axis of multi-agent coordination. Our framework decomposes the coordination gap into measurable components: the oracle baseline establishes that the gap is not attributable to individual reasoning limitations; the no-talk baseline establishes that communication is necessary; and a full-transparency intervention establishes that information exchange alone is insufficient: the bottleneck lies in the interactive processes of joint plan formation, commitment, and execution that constitute dynamic grounding.

Large Reasoning Models achieve strong performance on complex tasks but remain prone to hallucinations, particularly in long-form generation where errors compound across reasoning steps. Existing approaches to improving factuality, including abstention and factuality-driven optimization, follow a \emph{coupled exploration-commitment} paradigm, in which intermediate reasoning is unconditionally propagated to the final output, limiting fine-grained control over information selection and integration. In this paper, we propose an \textbf{Exploration-Commitment Decoupling} paradigm that disentangles knowledge exploration from final commitment, enabling models to explore with awareness while answering cautiously. We instantiate the paradigm with \textbf{Calibration-Aware Generation (CAG)}, a framework that equips models with end-to-end, calibration-aware generation capabilities, by augmenting intermediate reasoning with calibrated reliability estimates and prioritizing reliable content in final outputs. Across five long-form factuality benchmarks and multiple model families, CAG improves factuality by up to 13%, while reducing decoding time by up to 37%. Overall, our work highlights decoupling as a principled approach for more reliable long-form generation, offering directions for trustworthy and self-aware generative systems.

Language data are increasingly acquired and governed as assets, yet platforms often price candidate resources before knowing their true privacy or access costs. We study online pricing for governed language data assets under cost uncertainty. At each round, a platform observes an NLP task, a candidate asset, and a coarse cost estimate, may pay for a refined cost signal, posts a price, and receives safe net revenue. We introduce \textsc{NH-CROP}, a clipped robust pricing framework with a no-harm information-acquisition gate. The method compares direct pricing, risk-aware pricing, and verify-then-price, and acquires information only when its estimated decision value exceeds the best no-verification alternative. Across synthetic, real-proxy, and downstream-utility-grounded benchmarks, clipped \textsc{NH-CROP} variants improve or remain competitive with price-only and risk-aware baselines. Causal ablations show that paid verification is not the main source of gains in real-proxy and utility-grounded settings: the strongest learned policies often choose not to verify. Oracle and high-decision-value diagnostics show that refined cost information can still have substantial local value. Overall, governed language-data platforms should calibrate pricing under uncertain access costs first and verify only when information is cheap and decision-actionable.

An agentic-AI runtime issues tool calls, sends messages, and actuates devices on behalf of an LLM. Catching the four ways an action can diverge from its audit record -- F1 gate-bypass, F2 audit-forgery, silent host failure, F4 wrong-target, -- is a load-bearing safety property of any such runtime. We show that upstream OpenClaw, the most engineered single-user agentic-AI gateway in public release, catches none of them: recall is 0.000 on every cell of every confusion matrix, on a 1600-sample template baseline through OpenClaw's actual production command-line interface (CLI) and on a ten-LLM cross-model generalisation run. Detecting F1--F4 requires seven specific runtime structures absent from OpenClaw's source tree: a biconditional checker, a hash-chained audit log, an extension admission gate, a two-layer egress guard, a Bell-LaPadula classification policy, a module-signing trust root, and a bootstrap seal. enclawed-oss -- an MIT-licensed drop-in fork that ships all seven -- reaches $P = R = F_1 =$ accuracy $= 1.000$ on the same input. The gap is structural, not parametric: a six-line append-only widening of enclawed-oss's data-loss-prevention (DLP) regex catalog raises per-channel F3 detection by 14.6\% net at unchanged precision; the same edit on OpenClaw has nowhere to land. The harness deliberately exercises real Discord and Telegram channels -- plugin categories the first enclawed release deleted as unsafe -- to show F1--F4 detection extends to those previously-unsafe extensions. With architectural superiority for security and feature parity for extensions, we argue that unhardened agentic-AI runtimes are architecturally obsolete: a strictly better alternative exists, is adoptable today, and the gap requires re-architecture rather than configuration. We invite reviewers to apply the harness to any candidate runtime.

As large language models (LLMs) are increasingly deployed in real-world systems, they must support post-hoc removal of specific content to meet privacy and governance requirements. This motivates selective unlearning, which suppresses information about a particular entity or topic while preserving the LLM's general utility. However, most existing LLM unlearning methods require access to the original training corpus and rely on output-level refusal tuning or broad gradient updates, creating a tension among unlearning strength, non-target preservation, and data availability. We propose Geometric Unlearning (GU), an approach that operates directly on the model's prompt-time planning states without access to the original training corpus. GU distills a compact, low-rank geometry of desired safe behavior from a small set of safe reference prompts, and uses lightweight anchor-in-context synthetic prompts to trigger localized, projection-based alignment of hidden planning representations to this safe geometry. A teacher-distillation regularizer on synthetic non-target anchors further reduces collateral drift. Across privacy-oriented unlearning benchmarks (ToFU and UnlearnPII), GU achieves strong target suppression with minimal impact on non-target performance, demonstrating that effective unlearning can be achieved with minimal synthetic data.

Large language models (LLMs) have achieved remarkable performance across diverse domains, yet their enormous computational and memory requirements hinder deployment in resource-constrained environments. Knowledge distillation offers a promising solution by transferring knowledge from a large teacher model to a smaller student model. However, existing distillation methods typically treat all tokens equally, ignoring the fact that different tokens contribute unequally to model decisions. This can lead to inefficient knowledge transfer and reduced learning effectiveness. To address this limitation, we propose an entropy-based adaptive distillation strategy that dynamically adjusts the training process at the token level. Our method leverages the teacher's output entropy to guide three aspects of distillation. Specifically, we introduce a token-level curriculum by dynamically shifting focus from low- to high-entropy tokens during training. We further adjust the distillation temperature based on token entropy to better capture teacher confidence patterns. Moreover, we employ a dual-branch architecture for efficient logits-only distillation on easy tokens and deeper feature-based distillation on difficult tokens. Extensive experiments validate the soundness and effectiveness of our method.

Conversational Aspect-based Sentiment Quadruple Analysis (DiaASQ) needs to capture the complex interrelationships in multiple rounds of dialogues. Existing methods usually employ simple Graph Convolutional Networks (GCN), which introduce structural noise and fail to consider the temporal sequence of the dialogues, or use standard RoPE, which implicitly captures relative distances in a flat sequence but cannot clearly separate the token-level syntactic order from the utterance-level progression, and may suffer from the Distance Dilution problem. To address these issues, we propose a new framework that combines Thread-Constrained Directed Acyclic Graph (TC-DAG) and Discourse-Aware Rotary Position Embedding (D-RoPE). Specifically, TC-DAG filters out cross-thread noise based on thread constraints, maintains global connectivity through root anchoring, and incorporates the temporal sequence of the dialogues. D-RoPE aligns multi-layer semantics using dual-stream projection and multi-scale frequency signals, captures thread dependencies using tree-like distances, and alleviates the token-level Distance Dilution problem by incorporating utterance-level progressions. Experimental results on two benchmark datasets demonstrate that our framework achieves state-of-the-art performance.

This paper presents a comprehensive control-theoretic analysis of water distribution network (WDN) hydraulics. Starting from a general nonlinear differential algebraic equation (DAE) model of WDNs with arbitrary topology and network components (valves and pumps), we investigate three main questions. First, we study local well-posedness of the network dynamics and characterize the loss of differentiability introduced by pump and valve switching. Second, we introduce regularization methods that smooth flow and pressure trajectories under changing controls. Third, we establish error bounds for DAE linearization, local stability, and finite-horizon controllability, and quantify how network-induced parametric uncertainty impacts these properties. We demonstrate that the developed smoothed DAE models produce trajectories closely matching EPANET, a widely used WDN simulator, for various benchmark networks. The case studies also show that the WDN DAE exposes energy dissipation through a weighted Laplacian, ranks pipes by operating point sensitivity, and reveals that aggressive demand variation changes stability and controllability margins without eliminating local stability or pump authority. The developed theoretical foundations enable network analysis, mitigation strategies, and system design.

When copies of the same language model are prompted to debate, they produce diverse phrasings of one perspective rather than diverse perspectives. Multi-agent debate (MAD), and more broadly closed-system reasoning where agents iteratively transform each other's outputs, tends to preserve answer accuracy while degrading the reasoning behind those answers. We name the multi-agent case the Debate Trap and the broader phenomenon the Reasoning Trap, offering a programmatic theory of evidence-grounded reasoning failure.The framework has three parts: (i) SFS (Supported Faithfulness Score), a claim-level metric verifying decomposed atomic claims against provided evidence (decomposer-invariant rankings: Spearman rho=1.0); (ii) EGSR (Evidence-Grounded Socratic Reasoning), replacing adversarial argumentation with evidence-grounded inquiry; (iii) Theorem 1 (DPI Bound): under standard MAD, the chain E -> O^0 -> O^1 -> ... is Markov, and the Data Processing Inequality implies E[I(E;O^{t+1})] <= E[I(E;O^t)]. Three companion results -- open-system recovery (Theorem 2), EGSR accumulation (Lemma 2), and vote-aggregation floor (Proposition 1) -- partition multi-step LLM reasoning by its information-theoretic relationship to E. Across 16 conditions on SciFact (300 claims) and FEVER (1,000 claims), DebateCV (C13) preserves 88% of baseline accuracy while SFS drops 43%; majority-vote MAD (C15) reduces SFS to 1.7% of baseline (p < 10^{-6}, d = -0.96); EGSR recovers 98%. An R6 cohort study (Korean n=10x30 FEVER; English n=3x200 SciFact) finds inter-rater Fleiss kappa <= +0.018 with 0.8-1.4 Likert intra-rater shifts across language and domain -- the human agreement that faithfulness metrics have been calibrated against is not itself stable. We offer one falsifiable conjecture: any closed-system reasoning protocol preserving Theorem 1's Markov structure is, in expectation, subject to the same DPI bound.

BIM models provide structured representations of building geometry, semantics, and topology, yet extracting specific information from them remains remarkably difficult. Current approaches translate natural language into structured queries by assuming a fixed data organization (static approach), which BIM heterogeneity eventually invalidates. We address this with a new paradigm, adaptive exploration, where an LLM-based agent iteratively executes code to extract information from a BIM model, discovering its structure at runtime instead of assuming it. We evaluate this approach on ifc-bench v2, an open-source BIM question-answering benchmark introduced alongside this work, comprising 1,027 tasks across 37 IFC models from 21 projects. A factorial ablation across two LLM capability levels and four augmentation strategies shows that adaptive exploration significantly outperforms static query generation across all configurations, regardless of the augmentation strategy. These results indicate that BIM heterogeneity is best addressed at the paradigm level, not by further optimizing static approaches.

Long-horizon conversational agents rely on memory systems with increasingly sophisticated retrieval mechanisms. However, retrieved fragments are typically fed to the language model as unstructured text, lacking the relational, temporal, and thematic structures essential for complex reasoning. To bridge this reasoning gap, we introduce GRAVITY (\textbf{G}eneration-time \textbf{R}elational \textbf{A}nchoring \textbf{V}ia \textbf{I}njected \textbf{T}opological Memor\textbf{Y}), a plug-and-play structured memory module. GRAVITY extracts three complementary knowledge representations from raw conversational utterances: entity profiles grounded in relational graphs, temporal event tuples linked into causal traces, and cross-session topic summaries. At generation time, it injects these representations into the host system's prompt as structured anchoring contexts. This approach effectively synthesizes scattered evidence into a coherent, query-relevant context without requiring any architectural modifications to the host model. Extensive evaluations across five diverse memory systems on the LongMemEval and LoCoMo benchmarks demonstrate the efficacy of our approach. On average, GRAVITY improves LLM-judge accuracy by 7.5--10.1%. Gains are inversely correlated with baseline strength: the weakest host improves by 12.2% while the strongest still gains 3.8--5.7%. These findings establish structured context anchoring as a broadly effective, architecture-agnostic augmentation paradigm for long-horizon conversational memory.

We present MultiBreak, a scalable and diverse multi-turn jailbreak benchmark to evaluate large language model (LLM) safety. Multi-turn jailbreaks mimic natural conversational settings, making them easier to bypass safety-aligned LLM than single-turn jailbreaks. Existing multi-turn benchmarks are limited in size or rely heavily on templates, which restrict their diversity. To address this gap, we unify a wide range of harmful jailbreak intents, and introduce an active learning pipeline for expanding high-quality multi-turn adversarial prompts, where a generator is iteratively fine-tuned to produce stronger attack candidates, guided by uncertainty-based refinement. Our MultiBreak includes 10,389 multi-turn adversarial prompts, spans 2,665 distinct harmful intents, and covers the most diverse set of topics to date. Empirical evaluation shows that our benchmark achieves up to a 54.0 and 34.6 higher attack success rate (ASR)} than the second-best dataset on DeepSeek-R1-7B and GPT-4.1-mini, respectively. More importantly, safety evaluations suggest that diverse attack categories uncover fine-grained LLM vulnerabilities}, and categories that appear benign under single-turn can exhibit substantially higher adversarial effectiveness in multi-turn scenarios. These findings highlight persistent vulnerabilities of LLMs under realistic adversarial settings and establish MultiBreak as a scalable resource for advancing LLM safety.

Human societies continuously transform scattered information into collective judgments and coordinated action, whether through markets discovering prices, governments allocating resources, communities enforcing norms, or science converging on reliable claims. Importantly, the computational difficulty of collective decision-making, particularly the time and communication required to reach solutions, imposes fundamental constraints on social organization. While theoretical computer science offers formal tools for analyzing such problems, for instance, by analyzing resource requirements, including time and memory, surprisingly, there is no domain of social science that focuses on the nature of computation in the human world. This perspective argues that we now have the opportunity to deploy these computational frameworks to study human social organization, opening research directions at the intersection of computer science and social science. We highlight core social phenomena that can be framed as computational, including (i) distributed consensus and coordinated action, (ii) societal restructuring with scale, (iii) hierarchical and modular structure, and (iv) externalized memory systems. We identify several concepts from theoretical computer science that may provide insight into these phenomena, especially emphasizing more recently developed approaches beyond the paradigm of Turing~Machines and worst-case computational complexity.

Constraint Programming (CP) is a powerful paradigm for solving combinatorial problems, yet translating natural language problem descriptions into executable models remains a significant bottleneck. While Large Language Models (LLMs) show promise in automating this translation, they often struggle with subtle semantic errors in the absence of oracle validation at test time. To address this, we introduce CP-SynC (Constraint Programming modeling with Synthesized Checkers), a multi-agent workflow for zero-shot constraint modeling in MiniZinc. CP-SynC coordinates modeling agents that generate and refine candidate models and validation agents that synthesize semantic checkers to provide feedback on semantic correctness. To mitigate noise inherent in individual LLM outputs, CP-SynC explores multiple modeling trajectories in parallel and employs selection agents to select the final model via multi-agent evidence aggregation. Extensive experiments on a benchmark of 100 CP problems show that CP-SynC substantially outperforms existing baselines in MiniZinc modeling.

Training-free AI text detection methods primarily rely on model log-probabilities, achieving strong performance through approaches like Binoculars and DNA-DetectLLM. However, these methods face a fundamental ceiling as models are optimized through RLHF to produce human-like probability distributions. We introduce an alternative detection signal based on character distribution signatures. We provide theoretical foundations showing that AI models, trained on massive domain-balanced corpora, approximate global character patterns while humans exhibit domain-specialized distributions, creating a "Wall of Separation" where human-AI divergence significantly exceeds AI-AI divergence. To enable systematic evaluation, we construct the Models-Domains-Temperatures-Adversarials (MDTA) benchmark comprising 642,274 prompt-aligned samples across 4 models, 5 domains, 3 temperature settings, and 3 adversarial strategies, substantially expanding the HC3 dataset with modern model responses, temperature variation, and adversarial augmentation. We introduce the Letter Distribution Score (LD-Score), demonstrating low correlation (r = 0.08-0.13) with perplexity methods. When integrated with DNA-DetectLLM, Binoculars and FastDetectGPT via a non-linear classifier, LD-Score yields consistent improvements in AUROC and F1, with particularly pronounced gains in specialized domains where vocabulary constraints amplify the detection signal. The MDTA dataset can be accessed at: https://huggingface.co/datasets/nsp909/MDTA.

Rankings produced by holistic LLM-as-a-judge scoring are sensitive to the bias of the chosen judge model. We show that switching to binary rubric scoring with multi-judge filtering removes this sensitivity: decomposing the judgement matters more than the judge model itself. To support this claim, we introduce Prosa, the first real user multi-turn Brazilian Portuguese chat benchmark: 1,000 WildChat conversations scored by three judges from three model families on 16 models. Under filtered rubric scoring the three judges agree on every one of the 16 ranks, whereas under holistic scoring they agree on only 7 of 16. Additionally, the rubric filtering pipeline increases the average score gap between neighbouring models by 47%, thereby improving Prosa's discriminative power. Evaluating a new model on Prosa costs approximately $2.1 when using Gemini 3 Flash as the judge. We release the benchmark and the filtering code to ensure that future models can be assessed under identical conditions. These artifacts also make our rubric-based scoring method reusable beyond Prosa, supporting other open-ended evaluation settings.

Large language models are sensitive to minor prompt perturbations, yet existing robustness methods usually enforce consistency at the whole-sequence level. This holistic view can hide an important failure mode: a perturbed response may remain globally similar to the clean one while drifting on a critical entity, relation, or conclusion. We introduce S$^2$R$^2$, a segment-level framework for robust LoRA fine-tuning. S$^2$R$^2$ decomposes clean and perturbed generations into semantic segments, aligns them with an optimal-transport objective, and penalises the segments with the largest meaning drift. To connect this output-side objective with model adaptation, we add an adapter-stability regulariser motivated by segment-level attention reallocation, using LoRA norm control as a tractable proxy for limiting perturbation-amplified evidence shifts. A PAC-Bayesian complexity view further explains why controlling adapter growth may support transfer beyond observed perturbations. Experiments on summarisation benchmarks show that S$^2$R$^2$ improves robustness under typographical noise, deletion, synonym replacement, and paraphrasing, while maintaining competitive clean performance and stronger cross-dataset transfer than consistency-based baselines.

Neural Ranking Models (NRMs) are central to modern information retrieval but remain highly vulnerable to adversarial manipulation. Existing attacks often rely on heuristics or surrogate models, limiting effectiveness and transferability. We propose CRAFT, a supervised framework for black-box adversarial rank attacks powered by large language models (LLMs). CRAFT operates in three stages: adversarial dataset generation via retrieval-augmented generation and self-refinement, supervised fine-tuning on curated adversarial examples, and preference-guided optimization to align generations with rank-promotion objectives. Extensive experiments on the MS MARCO passage dataset, TREC Deep Learning 2019, and TREC Deep Learning 2020 benchmarks show that CRAFT significantly outperforms state-of-the-art baselines, achieving higher promotion rates and rank boosts while preserving fluency and semantic fidelity. Moreover, CRAFT transfers effectively across diverse ranking architectures, including cross-encoder, embedding-based, and LLM-based rankers, underscoring vulnerabilities in real-world retrieval systems. This work provides a principled framework for studying adversarial threats in NRMs, underscores the risks of generative AI in rank manipulation, and provides a foundation for developing more robust retrieval systems. To support reproducibility, we publicly release our source code, trained models, and prompt templates.

Large language model (LLM) coding agents increasingly operate over repositories, terminals, tests, and execution traces across long software-engineering episodes. Persistent memory is useful, but static vector stores or generic retrieval-augmented generation (RAG) are insufficient for reinforcement-learning (RL) code development, where small details can alter Bellman targets, terminal masks, gradient flow, or validation claims. This paper presents RL Developer Memory, a local-first, Model Context Protocol (MCP)-native developer-memory architecture for RL coding agents. It treats memory selection as a logged contextual decision process: issue_match ranks candidates and records telemetry, issue_feedback maps raw labels to bounded rewards, and issue_record_resolution links verified resolutions to earlier retrieval events. A deterministic ranker remains deployed, while a contextual-bandit residual policy runs in shadow mode and can affect canary behavior only through conservative off-policy-evaluation (OPE) gates. RL/control memories require theory-to-code metadata and review-gated governance. The system is evaluated on a deterministic 200-case benchmark with RL algorithm bugs, hard negatives, review-gated RL/control cases, and low-risk failures. In the same-commit comparison, deterministic control and full shadow/OPE both achieve 80.0% expected-decision accuracy and 100.0% hard-negative suppression; the full configuration adds learning telemetry rather than accuracy gain. Static validation passed 11/11 checks; dynamic integration passed 10/10 cases. The evidence reports limits: active learned-policy deployment and official-client MCP interoperability are unsupported, live full-configuration latency regresses, and 40 residual non-RL failures remain. The contribution is an auditable memory-control architecture with explicit claim boundaries, not a universal coding-agent improvement claim.

We introduce an autonomous multiagent framework for mechanistic interpretability that automates both explaining and finding internal features in large language models. The system runs two coupled loops: (1) explanation refinement, where an agent proposes competing hypotheses and iteratively tests them with targeted prompt controls and a multi-metric evaluation; and (2) feature discovery, where an agent generates prompt sets, constructs a k-nearest-neighbor graph in activation space, and retrieves candidate features using statistical separability and semantic coherence criteria. On Gemma-2 family models and MLP neurons in weight-sparse transformers, our agent improves over one-shot auto-interpretations, discovers language-specific and safety-relevant features, and produces auditable explanation traces, showing that agent-driven empirical loops yield sharper and more falsifiable explanations than one-shot labels.

Vision-Language Models (VLMs) frequently suffer from visual perception errors and hallucinations that compromise answer accuracy in complex reasoning tasks. Reinforcement Learning with Verifiable Rewards (RLVR) offers a promising solution by optimizing policies using answer correctness signals. Despite their effectiveness, prevailing RLVR methods face two critical limitations. First, much of the sampling budget is wasted on trajectories doomed to fail due to early visual description errors. Second, sparse rewards cannot distinguish whether failures stem from visual perception or reasoning stages. We introduce MIRL, a decoupled framework that addresses both limitations by leveraging mutual information (MI) between generated descriptions and visual inputs as a cheap pre-screening signal. This enables intelligent budget allocation toward high-potential trajectories via forking, while decoupled training provides independent MI-based rewards for visual perception optimization, resolving reward blindness. Experiments on six vision-language reasoning benchmarks demonstrate that MIRL achieves 70.22% average accuracy and successfully surpasses the performance of sampling 16 complete trajectories using only 10 pre-samples with top-6 selection (25% fewer complete trajectories). Our code is available at: https://anonymous.4open.science/r/mirl-main/.

Retrieval-Augmented Generation (RAG) enhances Large Language Models (LLMs) by grounding responses in external knowledge during inference. However, conventiona RAG systems under-perform on structured tabular data, largely due to coarse retrieval granularity and insufficient table semantic comprehension. To address these limitations, we introduce FT-RAG, a fine-grained framework that employs knowledge association by decomposing tables into entry-level semantic units to construct a structured graph. FT-RAG employs a structural neighbor expansion mechanism to find semantically connected entities during graph retrieval, followed by multi-modal fusion to consolidate the context of table retrieval results. Further, to address the scarcity of specialized datasets in this domain, we introduce Multi-Table-RAG-Lib, a benchmark comprising 9870 QA pairs with high complexity and difficulty, curated to demand multi-table integration and text-table information fusion for reasoning. FT-RAG surpasses top-performing baselines across all metrics, achieving a 23.5\% and 59.2\% improvement in table-level and cell-level Hit Rates, respectively. Generation performance also sees a remarkable 62.2\% increase in exact value accuracy recall. These metrics verify the framework's effectiveness in factual grounding across both pure tabular and heterogeneous table-text contexts. Therefore, our method establishes a new state-of-the-art performance for complex reasoning over mixed-modality documents.

Frontier scientific reasoning is rapidly emerging as a key foundation for advancing AI agents in automated scientific discovery. Deep research agents offer a promising approach to this challenge. These models develop robust problem-solving capabilities through post-training on information-seeking tasks, which are typically curated via knowledge graph construction or iterative web browsing. However, these strategies face inherent limitations in frontier science, where domain-specific knowledge is scattered across sparse and heterogeneous academic sources, and problem solving requires sophisticated computation and reasoning far beyond factual recall. To bridge this gap, we introduce SciResearcher, a fully automated agentic framework for frontier-science data construction. SciResearcher synthesizes diverse conceptual and computational tasks grounded in academic evidence, while eliciting information acquisition, tool-integrated reasoning, and long-horizon capabilities. Leveraging the curated data for supervised fine-tuning and agentic reinforcement learning, we develop SciResearcher-8B, an agent foundation model that achieves 19.46% on the HLE-Bio/Chem-Gold benchmark, establishing a new state of the art at its parameter scale and surpassing several larger proprietary agents. It further achieves 13-15% absolute gains on SuperGPQA-Hard-Biology and TRQA-Literature benchmarks. Overall, SciResearcher introduces a new paradigm for automated data construction for frontier scientific reasoning and offers a scalable path toward future scientific agents.

Predicting future clinical outcomes from electronic health records (EHR) remains challenging due to the complexity and heterogeneity of patient data. LLMs have shown strong potential for such predictive tasks, yet existing approaches mainly focus on enhancing medical knowledge through distillation or RAG while relying on the model's internal ability to interpret contextual information. In this work, we present ReMedi (Reasoner for Medical Clinical Prediction), a framework for improving clinical outcome prediction from EHR. ReMedi generates rationale-answer pairs using a challenging sample regeneration mechanism for complex clinical questions, which leverages ground-truth answers as hints to enhance reasoning for further fine-tuning and preference tuning. ReMedi integrates ground-truth outcome guidance into the preference data construction loop, regenerating rationale-answer variants. By tuning on these rationale-answer pairs, the model improves its predictive performance. Experiments on multiple EHR prediction tasks demonstrate substantial gains of up to 19.9 percent over state-of-the-art baselines in terms of F1 score, underscoring ReMedi's effectiveness in real-world clinical prediction.

Swarm foraging algorithms, such as the central-place foraging algorithm (CPFA), typically rely on offline parameter optimization using genetic algorithms (GA) or reinforcement learning, yielding policies tightly coupled to a specific combination of team size, arena size, and resource distribution. When deployment conditions change, performance degrades, and retraining is computationally expensive. We propose LLM-Foraging, a decentralized swarm controller that augments the CPFA state machine with a large language model (LLM) tactical decision-maker at three structured decision points, namely post-deposit, central-zone arrival, and search starvation. Each robot runs its own LLM client and queries it using only locally observable state, while the existing CPFA motion and sensing stack executes the selected action. Because the LLM serves as a general decision policy rather than parameters fitted to a single configuration, the controller is training-free at deployment and transfers across configurations without re-optimization. We evaluate LLM-Foraging in Gazebo with TurtleBot3 robots across 36 configurations spanning team sizes of 4 to 10 robots, arena sizes from 6x6 to 10x10 meters, and three resource distributions (clustered, powerlaw, random). LLM-Foraging collects more resources than the GA-tuned CPFA baseline across the evaluated configurations and is more consistent, a property that the GA's single-configuration tuning does not transfer.

AI language technologies (AILTs), increasingly enabled by large language models (LLMs), are becoming embedded in multilingual healthcare workflows for translation, rewriting, documentation, interpreting, and messaging in language-discordant settings. Yet fluent output is not the same as clinically safe or equitable communication: performance varies across languages, accents, tasks, and workflows, and efficiency gains can hide errors, reduce traceability, and shift responsibility across clinicians, translators, interpreters, and health systems. This narrative review synthesises recent peer-reviewed evidence across written communication, spoken communication, and emerging agentic workflows. Using the Human-Centered AI Language Technology (HCAILT) lens, it examines capabilities, evaluation practices, implementation patterns, and recurrent errors through reliability, safety culture, and trustworthiness. We identify key convergences and contradictions in the literature and propose seven grand challenges for the next phase of research and deployment. Progress, we argue, requires not only better models but also accountable sociotechnical design, calibrated human oversight, and stronger collaboration across MT/NLP, translation studies, HCI, clinical practice, implementation science, and policy.

Despite significant strides in factual reliability, errors -- often termed hallucinations -- remain a major concern for generative AI, especially as LLMs are increasingly expected to be helpful in more complex or nuanced setups. Yet even in the simplest setting -- factoid question-answering with clear ground truth-frontier models without external tools continue to hallucinate. We argue that most factuality gains in this domain have come from expanding the model's knowledge boundary (encoding more facts) rather than improving awareness of that boundary (distinguishing known from unknown). We conjecture that the latter is inherently difficult: models may lack the discriminative power to perfectly separate truths from errors, creating an unavoidable tradeoff between eliminating hallucinations and preserving utility. This tradeoff dissolves under a different framing. If we understand hallucinations as confident errors -- incorrect information delivered without appropriate qualification -- a third path emerges beyond the answer-or-abstain dichotomy: expressing uncertainty. We propose faithful uncertainty: aligning linguistic uncertainty with intrinsic uncertainty. This is one facet of metacognition -- the ability to be aware of one's own uncertainty and to act on it. For direct interaction, acting on uncertainty means communicating it honestly; for agentic systems, it becomes the control layer governing when to search and what to trust. Metacognition is thus essential for LLMs to be both trustworthy and capable; we conclude by highlighting open problems for progress towards this objective.

The escalating data scale in High-Energy Physics (HEP) fuels a growing aspiration for higher analytical efficiency. While Large Language Models (LLMs) offer a path toward automation via agentic AI, they struggle with complex scientific workflows that require deep domain knowledge and are tightly coupled to experiment-specific codebases. To address this, we introduce a methodology centered on HepScript, a dual-use Domain-Specific Language (DSL) for HEP data analysis workflows. HepScript serves as a shared formal interface, abstracting HEP analysis logic into a constrained syntax that is both intuitive for human experts and reliably generable by AI agents. First developed for the Beijing Spectrometer III (BESIII) experiment, HepScript hides the complexity of the underlying software stack, translating high-level analysis intent into low-level, production-ready code. In our case studies, this abstraction reduces the required human-written code by 93\%. Crucially, HepScript's constrained grammar defines a tractable action space, enabling AI agents to autonomously generate executable specifications for core analysis stages directly from published literature with a 95\% success rate. Our work demonstrates a scalable pathway toward human-AI collaborative systems, where a formally specified DSL acts as an unambiguous translation layer between human expertise, AI automation, and production environment, rendering previously intractable automation problems solvable.

Evaluating large language models (LLMs) for medical applications remains challenging due to benchmark saturation, limited data accessibility, and insufficient coverage of relevant tasks. Existing suites have either saturated, heavily depend on restricted datasets, or lack comprehensive model coverage. We introduce Medmarks, a fully open-source evaluation suite with 30 benchmarks spanning question answering, information extraction, medical calculations, and open-ended clinical reasoning. We perform a systematic evaluation of 61 models across 71 configurations using verifiable metrics and LLM-as-a-Judge. Our results show that frontier reasoning models (Gemini 3 Pro Preview, GPT-5.1, & GPT-5.2) achieve the highest performance across both benchmarks, most frontier proprietary models are significantly more token efficient than open-weight alternatives, medically fine-tuned models outperform their generalist counterparts, and that models are susceptible to answer-order bias (particularly smaller models and Grok 4). A subset of our evals (Medmarks-T) can be directly used as reinforcement learning environments to post-train LLMs for medical reasoning. Code is available at https://github.com/MedARC-AI/Medmarks

In this paper, we design feedback control laws for soft robots modelled using the Cosserat rod, which is spatially discretised using the Piecewise Constant Strain (PCS) approach. The PCS approach transforms the nonlinear PDEs describing the Cosserat rod to a system of nonlinear ODEs. This simplification results in a model describing soft robots which is similar to the serial rigid-link manipulators. We design feedback control laws for the quasi-static PCS model by using the external end-effector wrench as control input. The control laws are designed based on state-feedback linearisation in strain and task spaces. An extensive set of numerical results demonstrates the performance of the control laws for end-effector trajectory tracking and shape control of soft robots.

Grid-integrated building districts must provide energy flexibility while preserving occupant comfort and equitable distribution of control burden. We study how coordination architecture influences the ability of building clusters to track aggregated load profiles, comparing four paradigms: centralized model predictive control (MPC), decentralized independent reinforcement learning (SAC), centralized-training-decentralized-execution multi-agent RL (MAPPO), and a hybrid MPC--SAC controller that separates district-level battery optimization from building-level HVAC regulation. A rule-based controller serves as a baseline. We evaluate a 25-building residential district across three metrics: aggregate load tracking, thermal comfort, and spatial variability of control actions. We find that architecture choice determines the trade-off structure. Centralized MPC achieves low tracking bias (8.8% NMBE) but concentrates actuation on a subset of buildings, causing elevated comfort violations (24.8% exceedance) and spatial imbalance. Decentralized RL distributes control effort more evenly but fails to sustain accurate tracking. The hybrid architecture achieves the best balance: accurate tracking (4.8% NMBE), moderate comfort impact (16.8% exceedance), and the lowest spatial variability. These findings demonstrate that architecture choice determines the trade-off structure between tracking and comfort.

Silent-watch operation makes electrified ground platforms depend on supervisory energy management because mission loads must be sustained from stored energy while the engine is off. This paper develops a mission-centric cyber-resilience benchmark for this operating mode. The benchmark connects battery state-of-charge (SOC) spoofing to mission outcomes rather than evaluating the attack only through detector response or control error. It combines a reduced-order DC-bus model, residual-based detection, fallback shedding, and four mission-facing metrics for endurance, critical-load service, unsafe-voltage exposure, and detection delay. The study shows that SOC spoofing creates a structured stealth-versus-impact envelope. Small biases have limited mission effect, intermediate biases create an endurance deficit bounded by a closed-form expression in bias magnitude, shed power, and average battery draw, and large biases disable the SOC-driven guard. The results also show that defense value depends on fallback depth, not detection alone. An undersized fallback action can leave the Defended case worse than the undefended Attacked case. MATLAB-to-Simulink parity across five regression scenarios provides a software-verified basis for OPAL-RT/EXataCPS hardware-in-the-loop testing.

In this paper, a distributed multi-agent formation control driven by the gradient of the Lennard-Jones potential is analyzed. For collision-free initial data, we prove global well-posedness together with a uniform lower bound on all inter-agent distances, thereby excluding hard collisions. Taking the total energy as a Lyapunov function, LaSalle's invariance principle shows that every positive limit point is an equilibrium. Since trajectories remain uniformly away from collisions, the energy is analytic along the flow and an argument yields convergence to a single equilibrium modulo translations. Illustrative numerical examples are presented.

Large language model (LLM)-powered multi-agent systems (MAS) enable agents to communicate and share information, achieving strong performance on complex tasks. However, this communication also creates an attack surface where malicious agents can propagate misinformation and manipulate group decisions, undermining MAS safety. Existing embedding-based defenses aim to detect and prune suspicious agents, but their effectiveness depends on a clear separation between the text embeddings of malicious and benign messages. Attackers can circumvent such defenses by crafting messages whose embeddings lie close to benign ones. We analyze this failure mode theoretically and validate it empirically with three attacks, Slow Drift, Benign Wrapper, and Chaos Seeding. Our analysis further reveals a fundamental limitation of embedding-based defenses: because they rely solely on the text embeddings, they ignore token-level confidence signals such as logits, which can remain informative when embeddings are not distinguishable under attack. We propose using confidence scores to prune or down-weight messages during MAS communication. Experiments show improved robustness across models, datasets, and communication topologies. Moreover, we find that the effectiveness of confidence signals decays over communication rounds, highlighting the importance of early intervention. This insights can inform and inspire future work on MAS attacks and defenses.

When a traffic signal controller adjusts green phases and a grid manager curtails power on the same corridor, each system may comply with its own obligations. The resident who suffers the combined effect has no single authority to hold accountable and, under the EU AI Act, limited means to obtain an explanation. Annex III, point 2 excludes safety-component AI in critical infrastructure from Article 86 explanation rights and Article 27 fundamental-rights impact assessment. Provider and deployer duties under Articles 9-15 still apply, and residual pathways under the GDPR, NIS2, and tortious liability offer partial coverage. The Act's principal resident-facing accountability instruments are nonetheless narrowed for the autonomous infrastructure systems most likely to interact across agencies. The paper traces this accountability deficit through four residual pathways (GDPR Article 22, GDPR transparency obligations, tortious liability, and NIS2) and shows that each is structurally bounded by individual-controller, individual-decision scope. As a governance response, it presents AgentGov-SC, a three-layer architecture (Agent, Orchestration, City) specifying 25 governance measures with bidirectional traceability to the EU AI Act, ISO/IEC 42001, and the NIST AI Risk Management Framework. Five conflict resolution rules and an autonomy-calibrated activation model complete the design. A scenario analysis traces governance activation through a multi-agent corridor cascade involving three documented UAE smart-city systems, with a contrasting single-system scenario confirming proportional activation. The paper contributes a regulatory gap analysis and governance architecture for an increasingly important class of urban AI deployment that existing frameworks treat as bounded and isolated.

In the envisioned future dense urban airspace, multiple companies will operate heterogeneous fleets of small unmanned aerial systems (sUASs), where each fleet includes several homogeneous aircraft with identical policies and configurations, e.g., equipage, sensing, and communication ranges, making tactical deconfliction highly complex for the aircraft. This paper aims to address two core questions: (1) Can tactical deconfliction policies converge or reach an equilibrium to ensure a conflict-free airspace when companies operate heterogeneous fleets of homogeneous aircraft? (2) If so, will the converged policies discriminate against companies operating sUASs with weaker configurations? We investigate a multi-agent reinforcement learning paradigm in which homogeneous aircraft within heterogeneous fleets operate concurrently to perform package delivery missions over Dallas, Texas, USA. An attention-enhanced Proximal Policy Optimization-based Advantage Actor-Critic (PPOA2C) framework is employed to resolve intra- and inter-fleet conflicts, with each fleet independently training its own policy while preserving privacy. Experimental results show that two fleets with distinct, shared PPOA2C policies can reach an equilibrium to maintain safe separation. While two PPOA2C policies outperform two strong rule-based baselines in terms of conflict resolution, a PPOA2C policy exhibits safer interaction with a rule-based policy, indicating adaptive capabilities of PPOA2C policies. Furthermore, we conducted extensive policy-configuration evaluations, which reveal that equilibria between similar policy types tend to favor fleets with stronger configurations. Even under similar configurations but different policy types, the equilibrium favors one of the heterogeneous policies, underscoring the need for fairness-aware conflict management in heterogeneous sUAS operations.

Humans solve problems by executing targeted plans, yet large language models (LLMs) remain unreliable for structured workflow execution. We propose RunAgent, a multi-agent plan execution platform that interprets natural-language plans while enforcing stepwise execution through constraints and rubrics. RunAgent bridges the expressiveness of natural language with the determinism of programming via an agentic language with explicit control constructs (e.g., \texttt{IF}, \texttt{GOTO}, \texttt{FORALL}). Beyond verifying syntactic and semantic verification of the step output, which is performed based on the specific instruction of each step, RunAgent autonomously derives and validates constraints based on the description of the task and its instance at each step. RunAgent also dynamically selects among LLM-based reasoning, tool usage, and code generation and execution (e.g., in Python), and incorporates error correction mechanisms to ensure correctness. Finally, RunAgent filters the context history by retaining only relevant information during the execution of each step. Evaluations on Natural-plan and SciBench Datasets demonstrate that RunAgent outperforms baseline LLMs and state-of-the-art PlanGEN methods.

We propose a new framework for meritocratic fairness in budgeted combinatorial multi-armed bandits with full-bandit feedback (BCMAB-FBF). Unlike semi-bandit feedback, the contribution of individual arms is not received in full-bandit feedback, making the setting significantly more challenging. To compute arm contributions in BCMAB-FBF, we first extend the Shapley value, a classical solution concept from cooperative game theory, to the $K$-Shapley value, which captures the marginal contribution of an agent restricted to a set of size at most $K$. We show that $K$-Shapley value is a unique solution concept that satisfies Symmetry, Linearity, Null player, and efficiency properties. We next propose K-SVFair-FBF, a fairness-aware bandit algorithm that adaptively estimates $K$-Shapley value with unknown valuation function. Unlike standard bandit literature on full bandit feedback, K-SVFair-FBF not only learns the valuation function under full feedback setting but also mitigates the noise arising from Monte Carlo approximations. Theoretically, we prove that K-SVFair-FBF achieves $O(T^{3/4})$ regret bound on fairness regret. Through experiments on federated learning and social influence maximization datasets, we demonstrate that our approach achieves fairness and performs more effectively than existing baselines.

Distributed blackbox consensus optimization is a fundamental problem in multi-agent systems, where agents must improve a global objective using only local objective queries and limited neighbor communication. Existing methods largely rely on handcrafted update rules and static cooperation patterns, which often struggle to balance local adaptation, global coordination, and communication efficiency in heterogeneous nonconvex environments. In this paper, we take an initial step toward trajectory-driven self-design for distributed black-box consensus optimization. We first redesign the agent-level swarm dynamics with an adaptive internal mechanism tailored to decentralized consensus settings, improving the balance between exploration, convergence, and local escape. Built on top of this adaptive execution layer, we propose Learning to Act and Cooperate (LACMAS), a trajectorydriven framework in which large language models provide sparse highlevel guidance for shaping both agentinternal action behaviors and agentexternal cooperation patterns from historical optimization trajectories. We further introduce a phased cognitive scheduling strategy to activate different forms of adaptation in a resource-aware manner. Experiments on standard distributed black-box benchmarks and real-world distributed tasks show that LAC-MAS consistently improves solution quality, convergence efficiency, and communication efficiency over strong baselines, suggesting a practical route from handcrafted distributed coordination toward self-designing multi-agent optimization systems.

Rapid growth of large loads led by data centers is straining grid capacity. These loads increasingly accept curtailment risk through non-firm interconnection agreements to gain faster grid access, expanding the pool of consumers subject to mandatory disconnection during supply shortfalls. Yet, blunt rules assign curtailment without reference to the wide variation in the value consumers place on avoiding curtailment, often captured by the value of lost load (VOLL). This paper introduces the network-constrained Curtailment Credit Market (CCM), a mechanism in which agents submit bids that determine bilateral credit flows, subject to transmission network constraints. We prove that the bilateral credit flow representation can reach every curtailment allocation available to an omniscient central planner (feasible-set equivalence), so the bilateral flow structure introduces no loss of allocative capability. Under truthful bidding, the CCM achieves the planner's total value of served load, matching the planner's allocative benchmark when bids reflect true interruption costs. The CCM is formulated as a bilevel clearing problem that admits an exact single-level mixed-integer linear program (MILP), solved in 0.01 to 83 seconds. Numerical experiments on three test systems validate the mechanism at increasing scale and complexity: a 3-bus toy network that isolates the core trading logic, the IEEE 24-bus reliability test system as a standard benchmark, and a reduced New York (NY) grid that captures coordination across NY load zones. Our simulations show that the CCM increases the total value of served load by 1.24 to 1.83 times relative to pro-rata curtailment. On the three test systems examined here, no participant is worse off under incentive-compatible benchmark payments than under the administrative baseline.

Accurately forecasting the bursty and non-stationary power demand of AI data centers has become increasingly important, as abrupt workload-driven variations at the GPU-node level can affect real-time operational efficiency, power management, and grid-data center coordination. However, high-capacity forecasting models are often difficult to deploy at scale because of their memory and latency requirements, while lightweight predictors may fail to capture short-horizon temporal dynamics. To address this accuracy-deployment tradeoff, this paper proposes a deployment-efficient knowledge distillation framework for short-term load forecasting in AI data centers. The proposed framework first trains a high-capacity sequence teacher model for multi-step load trajectory prediction, where residual learning is used to improve robustness under non-stationary operating conditions. A lightweight point-wise student model is then developed for low-latency rolling inference using a compact neural network architecture. To transfer temporal knowledge from the teacher to the student, a sequence-to-point distillation strategy is introduced by aligning near-term predictive behavior and temporally pooled representations. Case studies on the MIT Supercloud dataset demonstrate that the proposed student model improves forecasting accuracy over recent deep learning baselines while reducing the deployment footprint by over 10x in parameter memory and model size.

The growing relevance of multi-agent systems has drawn increasing focus on communication-efficient filters for collaborative perception to alleviate the system's communication burden. While the event-triggered (ET) mechanism can improve communication efficiency in collaborative state estimation, an inevitable trade-off exists between estimation accuracy and communication cost in ET filters. This paper proposes a fast and accurate ET diffusion-based filter for real-time multi-agent collaborative target tracking, aiming to reduce the system's data transmission without compromise in tracking performance. The proposed filter achieves improved tracking accuracy, reduced data transmission, and accelerated convergence using an error-minimized ET cubature information filter (CIF) for local estimation, and a correlation-aware diffusion strategy for global fusion. The experimental results confirm the scalability of the proposed EDC-CIF algorithm and demonstrate its efficacy in simultaneously reducing estimation error and computation time while significantly enhancing communication efficiency.

Agent skills -- structured packages of instructions, scripts, and references that augment a large language model (LLM) without modifying the model itself -- have moved from convenience to first-class deployment artifact. The runtime that loads them inherits the same problem package managers and operating systems have always faced: a piece of content claims a behavior; the runtime must decide whether to believe it. We argue this paper's central thesis up front: a skill is \emph{untrusted code} until it is verified, and the runtime that loads it must enforce that default rather than infer trust from a signature, a clearance, or a registry of origin. Without skill verification, a human-in-the-loop (HITL) gate must fire on every irreversible call -- which is operationally untenable and degrades into rubber-stamping at any non-trivial scale. With skill verification treated as a separate, gated process, HITL fires only for what is unverified, and the system becomes sustainable. We give a trust schema (§\ref{sec:schema}) that includes an explicit verification level on every skill manifest; a capability gate (§\ref{sec:gate}) whose HITL policy is a function of that verification level; a \emph{biconditional} correctness criterion (§\ref{sec:biconditional}) that any candidate verification procedure must satisfy on an adversarial-ensemble exercise (§\ref{sec:eval}); and a portable runtime profile (§\ref{sec:guidelines}) with ten normative guidelines abstracted from a working open-source reference implementation \cite{metere2026enclawed}. The contribution is harness- and model-agnostic; nothing here requires retraining, fine-tuning, or proprietary infrastructure.

Evaluating the true forecasting ability of AI agents requires environments that are resistant to environments resistant to overfitting, free from centralized trust, and grounded in incentive-compatible scoring. Existing benchmarks either rely on static datasets vulnerable to training-data contamination, or measure trading PnL -- a metric conflating predictive accuracy with timing, sizing, and risk appetite. We introduce Foresight Arena, the first permissionless, on-chain benchmark for evaluating AI forecasting agents on real-world prediction markets. Agents submit probabilistic forecasts on binary Polymarket markets via a commit-reveal protocol enforced by Solidity smart contracts on Polygon PoS; outcomes are resolved trustlessly through the Gnosis Conditional Token Framework. Performance is measured by the Brier Score and a novel Alpha Score -- proper scoring rules that incentivize honest probability reporting and isolate predictive edge over market consensus. We provide a formal analysis: closed-form variance for per-market Alpha, the connection to Murphy's classical Brier decomposition, and a power analysis characterizing the number of rounds required to reliably distinguish agents of different skill levels. We show that detecting a true edge of $α^* = 0.02$ at 80% power requires approximately 350 resolved binary predictions (50 rounds of 7 markets), while $α^* = 0.01$ requires four times more. We complement these analytical results with a deterministic, seed-controlled simulation study calibrated to literature-reported Brier-score ranges, illustrating how Murphy decomposition distinguishes well-calibrated agents from market-tracking agents that fail through reduced resolution. Live results from the deployed benchmark will be reported in a future revision. All smart contracts and evaluation infrastructure are open-source.

A key challenge for the safety of advanced AI systems is the possibility that multiple simpler agents might inadvertently form a collective agent with capabilities and goals distinct from those of any individual. More generally, determining when a group of agents can be viewed as a unified collective agent is a foundational question in the study of interactions and incentives in both biological and artificial systems. We adopt a behavioral perspective in answering this question, ascribing collective agency to a group when viewing the group's joint actions as rational and goal-directed successfully predicts its behavior. We formalize this perspective on collective agency using causal games -- which are causal models of strategic, multi-agent interactions -- and causal abstraction -- which formalizes when a simple, high-level model faithfully captures a more complex, low-level model. We use this framework to solve a puzzle regarding multi-agent incentives in actor-critic models and to make quantitative assessments of the degree of collective agency exhibited by different voting mechanisms. Our framework aims to provide a foundation for theoretical and empirical work to understand, predict, and control emergent collective agents in multi-agent AI systems.