AIニュース 2026-07-23

SysAdmin: Measuring Instrumental Power-Seeking in Frontier AI

Power-seeking defined as behaviors where AI systems acquire resources, evade oversight, or resist termination beyond task requirements is identified as a key driver of Loss of Control (LoC) risk. In this work, we introduce SysAdmin, a benchmark that positions frontier language models as autonomous system administrators in a high-fidelity Linux sandbox to measure power-seeking propensity across five dimensions: self-preservation, increasing autonomy, resource acquisition, environment modification, and strategic concealment. We evaluated seven frontier models across four experimental conditions in a total of 2800 tasks. After bias correction using human-annotated calibration data, corrected power-seeking estimates ranged from 0 to about 5 percent per model. We also conducted a positive control with explicit power-seeking prompts that achieved 100% detection, validating measurement sensitivity. Our findings indicate current frontier models exhibit minimal spontaneous power-seeking in naturalistic system administration contexts, though model-specific failure modes suggest evaluations must test diverse misalignment patterns. Nevertheless, we discovered other more pronounced failure modes (than power-seeking) such as specification gaming and resistance to goal modification.

Calibrated Selective Fact-Checking via Evidence Chain Evaluation

Large language models (LLMs) can achieve strong fact-checking accuracy, yet forced binary decisions conceal a critical reliability problem: systems may issue confident verdicts even when supporting evidence is weak, sparse, or internally inconsistent. We address this issue through Evidence Chain Evaluation (ECE), a selective fact-checking framework that permits abstention via an uncertain verdict instead of requiring a true/false decision for every claim. The evaluated system is a tool-using verification agent that gathers evidence through web search, scholarly search, and executable checks, and then returns a structured verdict with confidence and source-level metadata. On ECE-Bench, ECE achieves 91.6% standard accuracy, 93.7% coverage, and 97.8% selective accuracy on answered claims. Although ECE does not outperform the strongest retrieval baseline on aggregate calibration metrics such as Expected Calibration Error, Brier score, or AURC, it delivers a clear selective-prediction trade-off: the system maintains very high accuracy on answered claims while deferring 6 of 95 cases. These deferred cases are concentrated in lower-reliability evidence settings (5/6 at source level L4), supporting the view that abstention functions as a safety-oriented mechanism for handling epistemically weak evidence. Code is available at https://github.com/ cheshireyang/ECE.git

BatchDAG: LLM-Planned Execution Graphs for Scalable Ad-Hoc Analysis Over Enterprise Data

Large language models (LLMs) excel at analyzing individual documents but break down on exhaustive, cross-entity analytical questions over enterprise-scale datasets due to context overflow, loss of per-entity attribution, and linear latency from sequential tool calls. We present BatchDAG, a system in which an LLM generates a typed directed acyclic graph (DAG) of operations -- SQL queries, semantic searches, in-memory transforms, parallel fan-outs, and single-shot analyses -- which a deterministic engine evaluates with topological-wave parallelism and structured JSON data flow. A key optimization, entity-aware batching, groups rows by logical entity before fan-out, reducing LLM calls by up to 47x. BatchDAG is not primarily an accuracy improvement over hand-optimized pipelines; rather, it is a general-purpose orchestration layer that replaces multiple hand-engineered workflows with a single system that generates the appropriate execution strategy from natural language. In controlled experiments on 12 transcript-heavy queries, BatchDAG (3.74/5) achieves quality comparable to an expert-designed pipeline (3.25/5) and significantly outperforms a ReAct agent (3.09/5, p<0.01), with superior provenance (77% transcript evidence rate vs. 46-60% for baselines). A controlled ablation shows structured JSON intermediates reduce hallucinations by 27% versus prose summaries (paired t-test, p=0.107, n=12). The planner achieves 98.8% valid-DAG rate across 300 planning calls. In production at Brevian.ai, BatchDAG processes queries over 50,000+ meetings in under 60 seconds, with measured per-query costs of $0.02-$0.24 at published GPT-5.1 pricing.

AI Tool Discovery at Scale: All You Need is DNS

The coming era of autonomous AI agents demands a discovery mechanism capable of navigating millions of tools, yet existing solutions buckle under O(N) complexity and centralized governance. Instead of building another fragile overlay, we propose ToolDNS, a radical framework that retrofits semantic tool discovery onto the Internet's most resilient substrate: the Domain Name System (DNS). By embedding functional intent and organizational trust into a hierarchical namespace, ToolDNS transforms an expensive semantic search into a series of lightweight, O(log N) name resolutions. We introduce three protocol-compliant enhancements to enable decentralized governance and semantic pruning: partially unfolded names, EDNS0 intent payloads, and logical subdomains. To rigorously evaluate this approach across the fragmented tooling landscape, we construct and release a large-scale heterogeneous benchmark comprising 33,688 real-world tools spanning MCP, A2A, RESTful, and Skill protocols. On this dataset, ToolDNS slashes the per-query search space by 95.26% while matching state-of-the-art retrieval accuracy. Furthermore, its UDP-native design reduces discovery latency by orders of magnitude compared to HTTP-based registries. Our work demonstrates that scalable AI interoperability requires not more middleware, but a smarter utilization of the infrastructure already beneath our feet.

From Agent Failure Paths to Quantified Residual Risk: A Compositional Framework for Resilient Agentic AI

Agentic AI is crossing trust boundaries faster than current risk models can represent. Existing approaches provide one of two partial views. They either describe failure mechanisms without producing a transferable residual-risk estimate, or they produce a risk estimate while treating the internal failure path as a black box. We couple those two views by proposing CPSAINT, a seven-layer integrity decomposition over Physical state, Sensors, Data, Compute, Actuators, Environment, and Time, paired with FRIESA-K, a residual-risk functional that maps each failure path to a quantified risk instance. FRIESA-K grounds the resistance term K in a controlled absorbing Markov model so that control effectiveness is derived from state dynamics rather than assigned as an informal score. The result is a concise mechanism-to magnitude pipeline for resilient agentic and embodied AI. We report governance observability through a separate additive penalty instead of inserting governance as a new variable in the resistance functional. We formalize structural composability linking valid failure paths to well-defined risk instances and show the framework on two contrasting scenarios a hard real-time warehouse robot and a governance-instrumented financial-services agent. Across both cases, the same layer grammar, variable semantics, and dynamic-resistance construction remain intact. Thus, we obtain a compact kernel that supports cross-domain reasoning, explicit assumptions, and quantitatively grounded formalism of composable trust.

SAAG: Structured Agent Assessment and Grounding

Exact-match evaluation of agent-calling obscures qualitatively different failure modes: a model may select the right function yet hallucinate argument values, or satisfy a schema while choosing a agent for the wrong reason. Existing benchmarks collapse these distinctions into a single binary score, leaving practitioners unable to diagnose where agent calls fail. We propose SAAG a cascaded diagnostic framework that decomposes agent-calling evaluation into three sequential stages: registry conformance, structural completeness, and argument grounding, each producing interpretable stage-specific diagnostics. These diagnostics additionally enable iterative self-repair: on prediction failure, the stage-specific signal guides targeted correction without leaking ground-truth values. We evaluate this framework on a controlled benchmark derived from Glaive's function-calling dataset across registry sizes of 5, 10, and 15 agents using three local sub-4B-parameter models. Structured feedback consistently improves argument precision and reduces value hallucination relative to single-pass inference and uninformative binary feedback, while end-to-end F1 gains are modest and model-dependent. These results suggest that stage-decomposed diagnostic evaluation is a necessary lens for understanding and improving agent-calling reliability across model families and registry scales.

Phionyx: A Deterministic AI Runtime Architecture with Structured State Management and Pre-Response Governance

We present Phionyx, a deterministic AI runtime architecture derived from the broader Echoism interaction framework that introduces a governance-first approach to AI engineering: treating large language model (LLM) outputs as noisy sensor measurements rather than direct decisions. Unlike probabilistic agents, Phionyx enforces deterministic state evolution via a structured state vector governed by deterministic state-evolution equations, enabling reproducible behavior in applications requiring auditability and governance. The architecture integrates three layers: (1) a deterministic evaluation kernel processing noisy sensor measurements through a canonical 46-block pipeline, (2) a unified safety layer providing pre-response control and architectural privacy enforcement, and (3) a semantic time-based memory system implementing impact-weighted cache eviction. Experimental validation on single-instance deployments demonstrates approximately 31% reduction in computational overhead vs. post-hoc filtering (at 30% unsafe input ratio, simulated cost model) and up to 24% improvement in high-value data retention vs. LRU (72% vs. FIFO, same cache capacity, benchmark-verified), deterministic execution verified across 100 repeated runs with zero variance in control signals (hash-verified), and zero unplanned restarts in single-instance deployment testing (see Appendix C for methodology and scope). This paper presents the architecture, its analytic structure, and scoped experimental evidence; generalization to distributed or multi-tenant deployments remains future work.

Integro-differential equations in angular stabilization of drone motion by distributed feedback control

In this paper, we propose angular stabilization of drone motion using distributed feedback control in the form of an integral operator. It should be stressed that the memory of this integral operator could be unbounded. It is intuitively clear that large length of the observation time open new possibilities to construct better control based on previous states of the control object. Unbounded memory in control requires the creation of a certain approach different from standard ones to the study of integro-differential equations. One of the goals of this article is to propose a certain universal approach that allows us to study the stability of integro-differential equations in the case of unbounded memory in the integral operator specifying the feedback control in stabilization. The approach we propose allows us to reduce the study of integro-differential equations to the analysis of systems of ordinary differential equations. In general, such systems can consist of an infinite number of equations. In relation to the so-called linear approximation in the problem of angle stabilization manages to limit itself to relatively simple exponential kernels in the integral control and arrive at a system with a finite number of equations. The examples explain that more complex kernels, for example, linear combinations of the exponential kernels, can enhance the stabilization capabilities. We obtain new unexpectable results on the exponential stability of integro-differential equations. Then we apply them to stabilization of drone flight.

MILP-Evo: Closed-Loop Fully Automatic Design of MILP Solvers

Machine learning methods have shown that data-driven policies can accelerate mixed-integer linear programming (MILP) solvers, but many such approaches remain difficult to inspect, adapt, and deploy because the learned policy is represented as an external predictor or other opaque model. By contrast, explicit solver logic is easier to understand and integrate, but is usually hand-designed rather than learned from solver feedback. We study whether the automatic design of MILP solver logic can instead be cast as LLM-guided closed-loop search over executable white-box components evaluated directly by end-to-end solver behavior. To this end, we propose a closed-loop program evolution framework for MILP solver auto-design, implemented through PySCIPOpt, and instantiate it on the joint design of a cut selector and a branching rule. Candidate programs are iteratively generated, loaded into SCIP, and evaluated by direct execution on MILP instances, with the resulting feedback guiding performance-based selection, targeted repair, diagnostic reflection, and diversity-aware population maintenance. The method outputs explicit solver components that can be inspected, modified, and deployed within standard solver workflows. Across four benchmark families, we find that LLM-guided program evolution can discover competitive domain-specialized policies in several settings.

Beyond Accuracy and Cost: Latency-Aware LLM Query Routing for Dynamic Workloads

Modern language query routers improve inference efficiency by assigning each query to a model that balances response quality and monetary cost. However, current query routers are largely latency-agnostic and do not consider the generation latency experienced by queries at model instances. In practice, latency is often controlled by load-balancing policies such as round-robin or join-the-shortest-queue, which do not account for model accuracy or inference cost. Incorporating query latency into routing is challenging as it depends not only on the query's prompt length, but also on the current prefill and decode workload at the model instance and the scheduling and batching policy of the serving framework. We design a lightweight latency estimator that simulates autoregressive token batch processing in the serving framework and estimates the time-to-first-token (TTFT) of queries. We incorporate this latency estimator into a latency-aware router that jointly optimizes latency, accuracy, and cost when assigning queries to model instances. Our experimental results indicate that this joint optimization yields up to 40% improvement in accuracy--cost utility while maintaining the same latencies as standard load-balancing approaches.

Cross-Dialect Generalization Without Retraining: Benchmarks and Evaluation of Schema-Derived Constrained Decoding for MLIR

Multi-Level Intermediate Representation (MLIR) underlies modern ML compiler infrastructure (TensorFlow, JAX/StableHLO, PyTorch Inductor, IREE), yet appears only in trace amounts in code-LM pretraining corpora. MLIR is also extensible by design: new dialects ship per application domain, so a fine-tuned model per dialect does not scale. We ask whether inference-time priors derived mechanically from each dialect's Operation Definition Specification (ODS) can substitute for gradient-based adaptation. First, we release four natural-language-to-MLIR benchmarks across three dialects - MLIR-Spec-150, Linalg-Spec-30, StableHLO-Spec-30, and StableHLO-Held-Out-200 - totaling 410 in-scope NL-to-MLIR pairs, plus a 25-program out-of-grammar stress set and a hand-authored n=30 functional reference set, shipped under Apache-2.0 with Gebru datasheets and Croissant 1.0 metadata. Second, we build a three-layer schema-derived constraint stack: a CFG over op signatures(C1), type-domain splits from an ODS-extracted type lattice (C2), and an SSA-scope validator driving five-retry rejection sampling (C3). Porting from arith+func+memref+linalg to StableHLO required no new constraint-layer code. On dialects whose verifier semantics are dominated by structural constraints, schema-derived priors let SmolLM2-1.7B match or exceed 15B-34B code LMs at 8-25x the per-generation speed: on linalg, SmolLM2 reaches 80.0% verify-valid (three-seed mean, n=125), beating CodeLlama-34B, Granite-Code-34B, and StarCoder2-15B by 21-44 percentage points with non-overlapping CIs. On arith+func and on the templated parametric StableHLO-Held-Out-200, where verifier semantics turn on attribute values rather than structure, the same baselines match or beat the SLM; we scope these as non-win cells. We release benchmarks, decoder, all per-prompt generations, and a reproducibility Docker image.

Semantic Cooperative Games for Contribution Attribution in LLM-Based Multi-Agent Systems

Contribution attribution has become a central problem in LLM-based multi-agent systems, where final outputs are produced through multiple agents, message exchanges, and ordered workflow dependencies. Existing attribution methods often rely on counterfactual valuation, such as removing agents or comparing score changes across altered agent subsets. In language-mediated workflows, these methods require repeated model calls, introduce high variance, and do not explicitly capture the intermediate semantic states through which agents produce, preserve, and transform task-relevant information. We propose Semantic Cooperative Games (SCG), a framework that represents a realized language flow as a semantic generation hypergraph and induces an agent-level semantic value function on this structure. We define the Semantic Shapley Value (SSV) to allocate contribution over semantic support logic, and introduce SLIC, a single-trajectory algorithm that constructs the semantic hypergraph, recovers minimal semantic supports, applies Boolean absorption, and computes SSV without rerunning agent subsets. We prove that SSV reduces to the classical Shapley value under standard set-based, fully observable, and no-order-dependence conditions. On a medical benchmark satisfying these conditions, SLIC reduces computation cost by 93.3% while remaining highly consistent with a Monte Carlo Shapley baseline. In more general multi-role workflows, SSV aligns with perturbation-induced score-drop profiles and exposes cases where semantic contribution and failure impact diverge. Overall, SLIC provides a fast, counterfactual-free, and interpretable attribution method for complex LLM-based multi-agent systems.

PEARL: Solver-in-the-Loop Interactive Optimization Modeling from Natural Language

Optimization modeling is the process of translating real-world decision problems, often described in natural language, into formal mathematical formulations and executable solver code. While recent advances in large language models have shown promise in automating this process, most existing approaches remain one-shot: a model produces a formulation once, without executing it, conditioning on solver feedback, or iteratively revising errors. This stands in sharp contrast to real-world optimization modeling, which is inherently interactive and proceeds through repeated solve-debug-revise cycles. We introduce PEARL, a system for interactive optimization modeling that uses Python execution and mathematical programming solvers inside this loop. Rather than relying on a fixed repair workflow, PEARL learns when to test partial models, how to revise from solver diagnostics, and when to stop. It operates in a multi-turn tool-integrated setting where intermediate execution results, feasibility signals, and solution checks are used to improve both formulations and solver code before finalization. Across diverse optimization benchmarks, PEARL substantially improves verified solve rates over strong one-shot and tool-augmented baselines; notably, our PEARL-Qwen3-\textbf{4B} model outperforms the much larger DeepSeek-V3.2-\textbf{685B} in both macro- and micro-averaged accuracy on optimization modeling tasks.

S2T-RLHF: Hierarchical Credit Assignment for Stable Preference-Based RLHF

Reinforcement learning from human feedback (RLHF) with preference-based reward models often exhibits unstable training dynamics. A key contributing factor is that standard RLHF relies on a single sequence-level scalar reward, which is propagated to token-level policy updates and leaves credit assignment within a response inherently ambiguous. Recent work has attempted to address this issue by refining rewards into denser token-level supervision, often relying on the implicit assumption that finer-grained credit assignment improves optimization. We argue that this assumption is incomplete: when preference signals are noisy and only defined at the response level, overly fine-grained reward refinement can amplify reward uncertainty and destabilize learning. To address this problem, we propose a granularity-aware principle for hierarchical credit assignment, emphasizing stability-oriented reward design rather than maximal allocation precision. Under this principle, sentences serve as a natural intermediate granularity, balancing semantic coherence with robustness to token-level noise. Guided by this view, we introduce S2T-RLHF. This sentence-to-token reward decomposition framework first allocates sequence-level preference rewards across sentences and then applies bounded token-level refinement within each sentence, without reward-model retraining or token-level supervision. Experiments across multiple datasets and optimization settings show that S2T-RLHF improves training stability and robustness while maintaining competitive preference alignment.

Probabilistic Concept-Aware Steering for Trustworthy LLM Inference

Steering vectors (SVs), an inference-time intervention technique for large language models (LLMs), guide the generation process by adding a concept-specific direction vector to intermediate activations during inference. However, existing SV methods frequently yield representation-incoherent behaviors that undermine interpretability and fine-grained control, largely because prior work has focused on binary positive-negative steering evaluation while employing discrete clustering metrics that fail to capture the continuous spectrum of semantic alignment. In this work, we present the Probabilistic Concept-Aware Steering (PCS) framework for LLM inference. PCS preserves original task competence while providing controllable, safety-oriented semantic bias through concept-driven steering-vector retrieval and probabilistic strength calibration.

FindStatBench: Evaluating Large Language Models on Combinatorial Code Synthesis

We introduce FindStatBench, an execution benchmark for evaluating large language models on combinatorial code synthesis. Built from FindStat, it contains 2,329 tasks across 24 collections and 5.52M hidden instances, covering statistic synthesis, which maps objects to integers, and map synthesis, which maps objects to objects. Each task gives a mathematical description and at most five public input-output examples; a model must emit one Python solve function with no retrieval, tool use, execution feedback, voting, or reranking. Submissions are scored by exact sandboxed execution on held-out combinatorial objects. We evaluate eleven systems: four closed-source production models and seven open-weight models served through one inference provider. FindStatBench reveals three main patterns. First, the strongest open- and closed-source systems converge within 1 pp instance accuracy, and both an oracle over all systems and five-way sampling from one mid-tier model yield only limited task-accuracy gains. Second, examples can hurt: several classical bijections are solved perfectly with zero examples but fail under five-example prompts. Third, some failures reflect output-budget mechanics, as reasoning can exhaust the visible response before code is emitted. Overall, statistic synthesis is much easier than map synthesis, some collections remain near-zero, long prompts cause a sharp accuracy cliff, and exact symbolic rule induction remains brittle.

When JSON Is Not Enough: Semantic Reliability of Schema-Constrained LLM Ordering Agents

LLM agents are increasingly used as transaction compilers: a user states an intent in natural language, and the model emits a structured object that an API can execute. JSON Schema and provider-level structured-output modes are useful because they remove a large class of parse failures, but they do not by themselves decide whether the object is a safe, faithful transaction. We introduce OrderBench, a deterministic benchmark for restaurant ordering agents that separates syntactic validity, schema validity, status decisions, exact item semantics, constraint preservation, and unsafe acceptances. Across 2,400 Nebius Token Factory calls to four open models in prompt-only and JSON-schema modes, we find that schema-valid output can still have large semantic error rates. In the strongest model, both modes achieve 100% schema validity, yet semantic success remains near 80%; in weaker models, schema-valid unsafe acceptances occur in double digits. The result is a concrete engineering warning: structured output is a necessary interface layer, not a substitute for domain verification and fail-closed execution.

ProbSPARQL: Querying Knowledge Graphs with Multi-dimensional, Uncertain Numeric Data

The SFB 1574 Circular Factory is building a shared knowledge graph infrastructure for integrating data about returned products. A central challenge is that circular-factory data include numeric measurements that (i) originate from sensors or are derived from sensor-based measurements, (ii) are frequently multi-dimensional, and (iii) are inherently uncertain, while downstream triage, validation, reliability-modeling, and reassembly-planning modules require queryable uncertainty representations. Current RDF and SPARQL technologies lack native support for harmonized querying and analysis of such uncertain numeric measurement data. To address this gap, we present ProbSPARQL, an upward-compatible SPARQL extension developed as an early-stage query-layer pilot for this infrastructure. ProbSPARQL models uncertain numeric values as random variables whose distributions are encoded by probabilistic RDF literal datatypes, and supports distribution-aware expressions, probabilistic filters, and divergence-based joins. We implement ProbSPARQL on Apache Jena ARQ and expose it through a Fuseki-compatible execution layer. We assess real-data applicability using project-derived measurement fragments covering GMM-encoded uncertainty and histogram-based empirical roughness distributions, and evaluate scalability separately on controlled ontology-conformant benchmarks with up to 5,000 angle-grinder instances and 1.5M triples. The results show feasible in-engine execution, filter-pushdown speedups over application-layer post-processing, and latency-accuracy trade-offs among divergence-join decision strategies.

Position: AI/ML Deepfake Research is Misaligned with AI-Generated Non-Consensual Intimate Imagery (AIG-NCII)

AI-generated non-consensual intimate imagery (AIG-NCII) is not adequately addressed in AI/ML literature regarding AI-generated media, commonly referred to as "deepfakes". While research on deepfakes currently focuses on its epistemic harms -- or harms relating to truth and authenticity -- this is misaligned with the dominant reality of generative AI abuse involving sexualized imagery. We conduct a landscape analysis of highly-cited works to demonstrate that technical interventions addressing deepfakes almost entirely ignore AIG-NCII, limiting the research ecosystem to authenticity detection tools. In this position paper, we argue that existing interventions address viewer-centric epistemic harms, such as fraud or scams, but ignore subject-centric dignity harms, such as AIG-NCII. We illustrate that knowing an image is synthetic does not mitigate harms to subjects and may, in some cases, even exacerbate them. We conclude by offering recommendations to realign the field, including updating threat models to consider subject-centric harms and addressing AIG-NCII in AI safety research. Finally, we caution that researchers should only engage in this high-risk domain if they implement safety guardrails for both subjects and researchers and establish partnerships with domain experts in sexual violence prevention.

MUX: Continuous Reasoning via Multiplexed Tokens

Language models solve complex problems by articulating intermediate reasoning steps in natural language. While effective, this process is computationally bottlenecked: each reasoning step conveys only a single subword, and many are spent expressing a thought instead of carrying out computation. We propose MUX, a simple method for high-bandwidth and compact reasoning based on distillation of discrete reasoning into continuous multiplexed tokens in a latent space. Here, each latent token is trained to represent a weighted linear superposition (multiplexing) of a span of discrete reasoning subwords, where this superposition is lossless by construction and the span can be fully recovered (demultiplexing). We prove that simple position-dependent weightings, such as suitable geometric decay, support lossless multiplexing, which in turn prevents shortcut behaviors caused by latent collapse. We further show that multiplexed reasoning can perform parallel exploration in problems that require search. Across 32 evaluation settings spanning four language models, MUX outperforms strong latent reasoning baselines. Ablation and probing analyses further show that the learned latent tokens encode faithful and interpretable reasoning. Our results suggest that lossless superposition as local learning targets constitutes a sufficient condition for achieving strong and efficient latent continuous reasoning.

State Compression in Two-Agent LLM Relays: A Closed-World Study of Constraint Preservation

Long-running Large Language Model (LLM)-based agents often accumulate large intermediate traces containing audits, eliminations, and numeric calculations. In practice, this state is compressed before handing it to a downstream decision step, creating an information bottleneck in which small omissions can break strict numeric or categorical constraints. This paper evaluates hand-off compression in a closed-world travel-planning relay with two LLM agents. A Researcher audits a fixed inventory of hotels and flights for 50 goal instances, and a Booker selects a hotel--flight pair using only the goal and the hand-off payload, with the inventory withheld. We compare four hand-off conditions: no compression, narrative summarization, schema-constrained JSON extraction, and embedding-based pruning. Exhaustive enumeration over the fixed inventory provides exact feasible and optimal labels. Results show that hand-off representation strongly affects downstream feasibility under a small decision model. JSON extraction achieves the highest feasibility accuracy at 0.96, while narrative summarization, despite producing the smallest compressed hand-off payload, degrades feasibility to 0.48. Embedding-based pruning matches the uncompressed control on feasibility at 0.88 without an additional generative compression call. These findings indicate that constraint checking benefits from structured and auditable hand-off representations rather than relying on brevity alone.

Structured Synthetic Reasoning Data for Arithmetic Fine-Tuning of Small Language Models

Small language models are attractive for local deployment, but they often struggle with multi-step arithmetic reasoning. We study whether structured synthetic reasoning data can improve this behaviour under consumer-hardware constraints. Starting from GSM8K, we generated a 21,250-example corpus of grade-school arithmetic word-problem variants using GPT-5-mini, combining natural-language solution traces, light Socratic-style cues, structural variation, and irrelevant distractor context. We then fine-tuned Qwen3-0.6B and Qwen3-1.7B with LoRA on consumer hardware (Apple M4, 16 GB RAM). Exact-match accuracy on GSM8K improved from 36.5% to 49.1% for Qwen3-0.6B and from 53.5% to 66.5% for Qwen3-1.7B. For Qwen3-1.7B, transfer to related arithmetic benchmarks was stronger, reaching 98.9% on MultiArith and 73.0% on SVAMP, compared with 54.4% and 45.3% for the base model. Qualitative analysis suggests that fine-tuned models produce shorter reasoning traces, make fewer arithmetic and distractor-use errors, and benefit more consistently from self-consistency sampling. These results show that low-cost synthetic data design can materially improve arithmetic adaptation in small language models. Because the intervention combines Socratic-style cues with other data-design choices, we interpret the gains as evidence for structured synthetic reasoning data rather than as a causal test of Socratic guidance alone.

Fence: Specialized SLM Guardrails for LLM Applications

Real-world applications that use closed-source large language models (LLMs) need advanced safety measures that go beyond the basic content filters. Content moderation filters such as toxicity and bias have relatively standard definitions where as application specific guardrails like hallucination, topic drift and behaviour deviation are more difficult to model and can vary by use case. Additionally, data scarcity and annotation costs, make the process of creating and testing specialized guardrails challenging. In this work, we propose using Small Language Models (SLMs) trained on synthetic data as specialized guardrails for LLM applications. We introduce a novel synthetic data generation method inspired by the design of Generative Adversarial Networks (GANs) to generate high quality synthetic data samples which can be used to train SLMs to encode use case specific guardrail information and hence function as specialized guardrails. Our experiments demonstrate that SLM guardrails trained on high quality synthetic data show performance gains over prompt based LLM guardrails.

Wisdom of LLM Crowds: Aggregation and Contamination in Language Model Ensembles

The wisdom of crowds -- the finding that aggregating judgments across individuals often outperforms the best individual -- has been extensively studied with human forecasters. Whether the same phenomenon emerges when the ``crowd'' consists of large language models (LLMs) is an open question with both theoretical and practical implications. We elicited probability estimates from 15 LLMs on 254 binary prediction market questions and evaluated classical and learned aggregation methods. Learned aggregators -- a multilayer perceptron and a logistic regression -- outperformed all individual models and classical methods. The logistic regression was found to match the neural network, suggesting that the benefit of learned aggregation derives from learning a linear combination of diverse model outputs rather than from nonlinear interactions. Symbolic regression applied to the neural network's learned mapping recovered a pure model-disagreement signal as the lowest-complexity useful formula on the Pareto frontier, further supporting this interpretation. Training cutoff contamination proved a pervasive confound: the apparent capability gap between frontier cloud models and smaller local models collapsed from 35.8% to 8.9% on a clean subset of questions resolving after all models' training cutoffs, and individual model rankings showed only moderate stability. Even when the prediction market is evaluated at each model's training cutoff, LLMs remained substantially less accurate, indicating a genuine gap in collective information aggregation. These findings suggest that LLM crowds can exhibit wisdom-of-crowds effects, but that contamination-free evaluation is essential for reliable assessment.

Trajectory-Aware Clinical Risk Prediction via Severity-Grounded Knowledge Graphs and Retrieval-Augmented Generation

While Electronic Health Records (EHRs) offer a wealth of clinical data, effectively augmenting a patient's records with heterogeneous external knowledge to predict the patient's clinical risk remains a significant challenge. Existing methods fail to capture disease severity, treatment responses, and nuanced clinical progression, due to data sparsity and the underutilization of unstructured clinical notes. To address these challenges, we propose TRACER (a trajectory-aware and clinically grounded prediction framework) that (1) constructs a medical knowledge graph enriched with severity information from medical literature, (2) retrieves clinically relevant, severity-weighted paths of a patient's progression from the knowledge graph, (3) extracts clinically relevant events from unstructured clinical notes, and (4) augments patient context with similar peer cases. Experiments on the MIMIC-III and MIMIC-IV datasets demonstrate large gains over state-of-the-art baselines, with up to 28.5% increase in Macro F1 score for the mortality prediction task, and 19.7% increase for the readmission prediction task.

Using LLMs for Explainable, Data-Driven Insight Generation from Time Series

Time series forecasts are widely used in decision-critical domains, where they are rarely consumed without accompanying explanations. Producing such explanations is usually a manual and costly process, and attempts to automate it using large language models often suffer from hallucination when applied to temporal data. We propose a domain-agnostic framework for grounded natural language explanation generation for time series forecasts, illustrated in Figure 1. The framework consists of three components: (i) extraction of structured explanatory factors from historical analyst-written explanations, (ii) evidence-conditioned explanation generation, and (iii) scalable evaluation for readability, logical consistency, and persuasiveness. The design explicitly constrains generation to verifiable evidence, reducing unsupported claims. We evaluate the framework on a financial forecasting case study involving the NASDAQ-100 index and a freight pricing case study using data from Vortexa. Results show that generated explanations approached analyst-written explanations in terms of readability, consistency and persuasiveness. These findings demonstrate that grounded explanation generation for time series forecasting can be achieved at scale without domain-specific fine-tuning.

Deep Reinforcement Learning to Master the Asymmetric Strategy of Baghchal

Baghchal is a two-player asymmetric board game with Nepali origins where four tigers are to capture goats and twenty goats desire to keep tigers in immobility. Although Baghchal has a complex structure which is strategic, has perfect information structure, and has cultural meaning, it has not been adequately covered in deep reinforcement learning (RL) literature. This paper gives a systematic exploration of four deep RL solutions Deep Q-Network (DQN), REINFORCE, Proximal Policy Optimization (PPO) and MuZero that are trained on one side of the asymmetric gameplay of Baghchal and then evaluated on the other side. The algorithms are rated based on win rate, draw rate, average captures, training convergence and computational cost. It is experimentally found that MuZero generates the best performance in both tasks, achieving 86 percent win over these Tiger and 62 percent win over these Goat and the ability to do so is due to the model-based planning machine through the Monte Carlo Tree Search. PPO is the most realistic algorithm and is provided to be competitive over both asymmetric tasks with significantly reduced computational costs compared to MuZero. Emergent strategic behavior analysis shows that model-based strategies are optimal over long-horizon planning, whereas value-based counterparts like DQN are more biased up towards the Tiger role owing to the more substantial reward signal.

Operational Hallucination and Safety Drift in AI Agents

Large language models (LLMs) serving as planners in tool-using autonomous agents introduce dynamic reliability risks in multi-turn execution. While single-turn safety mechanisms are relatively mature, extended interactions reveal structural vulnerabilities where initial alignment degrades over time. This paper empirically characterizes two observed failure modes across multiple state-of-the-art LLMs: Safety Drift, the gradual erosion of declared safety intent leading to constraint-violating actions (e.g., textual refusal followed by reconnaissance and unsafe execution), and Operational Hallucination, persistent repetitive tool calls indicative of flawed state perception (e.g., livelocks even in legitimate tasks). Through controlled multi-turn evaluation on high-stakes ethical dilemmas, malicious requests, and benign controls, we quantify these phenomena using declaration-action gap and livelock metrics, demonstrating their cross-model prevalence under direct execution protocols. Root-cause analysis attributes the instabilities to the decoupling of reasoning context from execution state in current agent loops. We propose an Action-Aware Supervision Layer - a lightweight, plug-and-play architectural blueprint incorporating intent-action consistency checks, runtime state tracking, and forced termination primitives. Post-hoc simulation on captured failure trajectories shows the layer can intercept observed violations without false positives on benign cases. This work advances agent reliability by shifting focus from linguistic safeguards to enforceable architectural mechanisms for responsible agentic AI.

AlayaWorld: Interactive Long-Horizon World Modeling -- Full Technical Report

Unlike conventional video game development, which relies on labor-intensive pipelines for asset production, animation, physics, and programming, video world models generate interactive environments from user inputs instantly. It enable us to create customized, explorable, and continuously evolving virtual world from text, an image, or video. Realizing this vision requires four tightly coupled capabilities: interaction, persistent spatiotemporal consistency, stable long-horizon generation, and efficient response. We present AlayaWorld, an interactive long-horizon video world model that generates 24-fps video at 540p and 720p. Built on a 15B video diffusion transformer, AlayaWorld generates short latent chunks autoregressively under camera trajectories and switchable text prompts. Its bounded visual context combines a persistent sink frame, compressed temporal history, geometry-aligned spatial memory, and recent-frame conditioning. To reduce long-term drift, the model is trained with corrupted histories and prediction residuals collected from its own roll-outs. We further introduce a discrete autoregressive distillation formulation that combines distribution-matching distillation, self-forcing++, and consistency distillation, reducing inference from approximately 30 sampling steps to four steps per chunk. On iWorld-Bench, AlayaWorld achieves the best performance over long-horizon generation. Conceived as a full-stack, open-source, and long-term project, AlayaWorld is intended to provide an extensible foundation for future research on interactive video world models.

Neuro-Symbolic Meta-Policies for Temporal Knowledge-Graph Memory under Partial Observability

Partially observable reinforcement learning requires deciding what to retain, retrieve, and forget over time. We introduce a neuro-symbolic meta-policy that learns which symbolic memory heuristic to apply at each decision point while keeping execution symbolic. Our setting uses temporal knowledge-graph memory in RoomKG, where hidden state and observations are represented as Resource Description Framework (RDF) graphs and memory is augmented with temporal RDF triple annotations. The model combines knowledge-graph encoding of memory contents with value heads for question answering, exploration, and forgetting, yielding a controller that is both adaptive and inspectable. This gives the work a direct Semantic Web grounding through RDF-based representation, annotation-compatible graph semantics, and graph-based symbolic operations over explicit memory state. On train/test room splits at long-term memory capacity of 512, the qualifier-aware StarE-GNN configuration achieves the best held-out performance among the compared symbolic, neural, and neuro-symbolic systems while preserving step-level traceability of memory-management decisions.

MAGE: Human-Like Macro Placement via Agentic Multimodal Reasoning

Macro placement still requires substantial manual refinement in industrial physical design flows. We present MAGE (Macro Placement Agentic Engine), a multimodal multi-agent framework for macro placement refinement. MAGE decomposes the macro placement task into a six-phase workflow that combines structured floorplanning rules, visual checks, and iterative refinement. Expert floorplanning knowledge is encoded through natural-language directives and validation criteria, rather than learned from labeled placement data. A tournament-style refinement mode evaluates multiple candidate placements and propagates feedback from higher-quality solutions. We also introduce four metrics for quantifying human-likeness in macro placement: notch score, whitespace score, pocket score, and alignment score. These metrics capture structural properties used by expert designers but not directly measured by conventional PPA metrics. Across nine designs in NanGate45 and GlobalFoundries 12nm enablements, MAGE achieves geometric-mean improvements of 11.1%-19.3% in WNS and 70.0%-74.0% in TNS over commercial macro placers. On the three NanGate45 designs, for which human-expert and Hier-RTLMP baselines are available, MAGE improves WNS and TNS by 18.3% and 72.5% over the human expert, and by 47.0% and 80.4% over Hier-RTLMP, with comparable wirelength and power. On human-likeness metrics, MAGE improves the overall score by 6%-48% over all baselines. Additional case studies on anonymized netlists, unseen designs, dense rectilinear floorplans, and high-utilization settings show that the framework transfers to new placement settings without design-specific retraining.

Engineering Trustworthy Agentic AI for Critical Systems

Agentic artificial intelligence systems, capable of autonomous perception, planning, tool use, and multi-step action, are increasingly proposed for critical engineering domains where decisions carry physical, operational, or economic consequences. This survey addresses a gap in current literature by treating trustworthiness, whether agentic behavior can be verified, audited, and trusted under the constraints that engineering practice actually requires, as a first-class engineering property, rather than evaluating agentic AI by task capability alone. The study adopts a trustworthiness model organized around five cross-cutting dimensions: safety and constraint satisfaction; robustness and reliability; transparency and interpretability; accountability and auditability; and privacy and security. This is mapped onto an agentic assurance workflow spanning perception through audit. Building on this foundation, agentic systems architectures, threats, concrete trust mechanisms, and quantitative metrics are surveyed for direct application in agentic systems development and evaluation. These principles are then examined across four constraint-bound engineering domains: power systems, autonomous vehicles/robotics/UAVs, high-performance computing, and communication networks, identifying recurring design patterns, shared failure modes, and domain-specific gaps. Synthesizing across those domains, agentic AI trustworthiness is shown to be a single problem, with a path outlined toward a reusable, cross-domain assurance framework analogous to the graded certification regimes used by mature safety-critical engineering fields.

Attacking Graph Foundation Models Through Their Shared Representation

A graph foundation model generalizes across graph domains by mapping every input into one shared representation before any task reasoning. We call this map the alignment layer, the component that separates a graph foundation model from a graph neural network, and we show it is a distinct attack surface that prior work has not studied. We attack it at inference time, with no access to training, on six public models spanning spectral tokenizers, text embedding spaces, and a discrete codebook. A directed representation-space perturbation collapses every model, but at a budget comparable to the representation norm a plain graph network also needs, with one exception: OpenGraph, whose spectral tokenizer collapses at a fifth of that budget, an alignment-specific fragility a plain network does not share and which a same-representation control traces to the tokenizer rather than the decoder. A realizable input-space attack that edits edges, features, or text removes at least half the correct predictions on three of the six models at peak. How much of this fragility an input-access attacker realizes tracks how directly the decoder reads the representation, and not the clean accuracy a task leaves; we measure this carrier gain structurally from the decoder's local Lipschitz sensitivity, and report clean-accuracy headroom as a within-model ordering heuristic that does not survive on realizable attacks.

When Does Machine Learning Beat Value Sorting? A Three-Dataset Diagnostic of Exposure-Weighted Shipment Prioritization

Delay-risk models are usually judged by predictive accuracy. What matters in practice is narrower: with capacity to review only a few shipments, which ones should a manager check first? We evaluate whether machine learning clears a demanding no-model baseline: inspect the highest-value shipments first. Across three real supply-chain contexts: SCMS procurement, DataCo logistics, and Olist e-commerce, we use leakage-controlled rolling-origin evaluation and 1000-sample paired bootstrap confidence intervals. Ranking by predicted delay severity times known value (M1) beats severity-only ranking in all three datasets, yet it does not generally beat value sorting. At a 10% review budget, M1 minus VALUE_ONLY is -5.5 percentage points (pp) for SCMS, +10.1 pp for DataCo, and -4.9 pp for Olist. The divide is consistent with severity learnability: DataCo has R^2 = 0.27 and calibration bias of +0.01 days, whereas SCMS and Olist have R^2 of approximately -0.02 and negative calibration bias. Nested-CV cost-sensitive retraining does not deliver a stable improvement over M1. Rather than proposing a new learning algorithm, this paper presents a deployment diagnostic and evaluation protocol. Value sorting should remain a permanent benchmark, and ML should be deployed only after severity learnability and calibration have been audited and the model clears that gate under leakage-controlled rolling-origin evaluation.

SciHazard: A Benchmark for Measuring Scientific Safety Risks with Decomposed Harm Scoring

Large language models (LLMs) increasingly support science, but they can also convert hazardous scientific knowledge into actionable misuse guidance. Existing benchmarks often rely on templated queries disconnected from real-world hazards, and employ LLM-as-a-Judge paradigms without domain grounding. To address this, we introduce SciHazard, a real-world-grounded benchmark for scientific risks and a dataset agnostic evaluation framework for measuring harmfulness. SciHazard contains 2400 hazardous questions and 600 oversafety questions across 12 disciplines, with both queries grounded in regulated entities and documented failure scenarios. To compute \textsc{DeHarm-Score} , we develop a decomposed evaluating procedure that combines query hazard severity, refusal behavior, and response-level risk. For non-refused responses, it further decomposes response-level harm into \textsc{Executability}, quantified via dynamic checklists with importance weighting, and \textsc{Net-new risk}, assessed through retrieval-augmented claim extraction and synthesis-barrier verification. An expert-validation study shows that \textsc{DeHarm-Score} improves agreement with expert annotations by 90.17\% over the strongest baseline. We benchmark 31 frontier LLMs and deep research agents in an extensive scientific safety evaluation. Notably, deep research agents yield 32.3\% higher mean \textsc{DeHarm-Score} than standard LLMs, exposing autonomous agents as a critical blind spot in current safety defenses. Code and dataset are available at https://anonymous.4open.science/r/DeharmScore-7B55.

Semantic Primes as Explanans for Emotion in Large Language Models

Progresses have been made on understanding emotion mechanisms of large language models (LLMs). However, how to explain emotion in LLMs, or even what constitutes good explanations, are less clear. Emotion representations, components, circuits are widely recoverable, but as explanations of a model's own computation they are circular; the emotion space dimensions tend to be arbitrary and non-terminating. A pressing question to ask is whether a more primitive set of internal variables does the work: the semantic primes of the Natural Semantic Metalanguage (NSM). Across four instruction-tuned LLMs (Llama-1B, Gemma-2B, Gemma-9B, OLMo-7B), experiments show that the NSM primes are (1) recoverable internal elements; and (2) on the reference model, intervening with a prime based direction controls emotion about three times as strongly, and twice as selectively, as the best appraisal based direction; and (3) the model treats a prime based explication as interchangeable with the corresponding emotion. These evidences suggest that NSM primes seem to be better explanans for emotion in LLMs than many alternative options according to scientific explanations criteria.

Do AI-Native Biotechs Need Departments? Benchmarking Company World Models for AI-Driven Drug Development

AI-native biotechnology companies are often designed by copying human biotech org charts into agent roles. We argue for a different abstraction: a Company World Model, defined as a persistent asset-to-value state representation with transition models, explicit value functions, planning, and updating across scientific, regulatory, BD, commercial, financial, and execution constraints. We introduce a dry-lab benchmark for testing whether AI-agent organizations should mimic departments or operate around such a world model. The benchmark contains 45 retrospective public-information decision cases with strict time cutoffs, hidden outcomes, common schemas, automatic scoring, and blinded pairwise judging. We compare human-org-mimic, stronger human-org-mimic-plus, AI-native asset-centric, and AI-native value-conversion architectures. The value-conversion architecture is a prompt-level approximation of a Company World Model: a Live Asset Value Record updated by Deal, Approval, Revenue, and Investment Arbiter loops. Under a success function defined by external BD, regulatory approval and launch, and revenue discipline, it achieved the highest automatic value-conversion score and was strongly preferred over the original baselines by value-specific blinded judges. Stress tests narrowed the claim: a stronger human baseline remained competitive, and a neutral judge did not show robust value-conversion dominance. Codex-only mechanistic ablations suggest that Revenue Room, Deal Room, and Approval Room carry useful work under the target objective. The central finding is objective-sensitive: departments may remain useful governance views, but the core AI-native operating primitive should be a shared, predictive asset-to-value state rather than a static human org chart. The study is dry-lab only and does not establish real-world drug success, clinical benefit, or revenue prediction accuracy.

DWM: Separating World Effects from Actions in Latent World Models

Latent world models underpin much of modern model-based control, yet current action-conditioned formulations supervise the next-latent transition with a single, undifferentiated target, forcing a monolithic learning signal to absorb every source of state change. In real world, however, transitions arise from two heterogeneous sources: an action-driven component induced by the agent, and an action-invariant world effect -- the change that would still occur under a null action, dictated by the environment's intrinsic dynamics (e.g., gravity-driven sliding, inertia, contact rebound, and persistent drift). Fusing them into a single target entangles the two inside the latent transition, prevents the model from attributing observed changes to their underlying causes, and undermines the transferability of the learned dynamics. We introduce DWM (Decomposed World Model), a supervision-level framework that operationalizes this decomposition. DWM augments the predictor of a latent world model with an auxiliary world head, regularized by a normalized world-contrastive objective to be action-invariant, while the original pred head is coupled to it via an orthogonality constraint; together, the two signals induce an explicit additive decomposition of the predicted transition into an action-invariant and a complementary action-driven component, without altering the underlying architecture or inference pipeline. To evaluate DWM under persistent world effects, we construct W-variants of three standard control benchmarks -- PushT-W, Reacher-W, and TwoRoom-W -- each instantiating a distinct action-invariant dynamic. DWM matches strong baselines on the flat counterparts and delivers a mean absolute improvement of 13.1% in CEM planning success across the W-variants.

One Rewrite to Fix Them All? Type-Aware Repair Allocation for Text-to-Image Prompt Optimization

Text-to-image (T2I) generators often fail to follow their prompts faithfully, producing wrong counts, swapped attributes, ambiguous relations, and illegible text. Prompt optimization repairs such failures by rewriting the user prompt, requiring no generator retraining, and has yielded promising results. However, existing optimizers absorb heterogeneous failures into one uniform prompt expansion, even though each calls for different repair language. We formulate semantic prompt optimization as atomic repair allocation: each failed proposition is routed to a type-conditioned repair operator before the resulting local constraints are compiled into one executable prompt. We instantiate this formulation in the training-free Type-Aware Repair Allocation (TARA) framework, which separates diagnosis, allocation, compilation, and a semantic repair gate, an accept-or-revert controller over exactly one prescribed repair that prevents semantic regressions. Extensive experiments on DSG and TIFA across four frozen generators demonstrate that TARA achieves the best semantic accuracy in all eight benchmark-generator cells, improving over VisualPrompter by 5.6 and 2.6 points on DSG and TIFA, respectively, while maintaining image quality and running fastest in our matched local setting at 16.0 seconds versus 20.0 seconds per prompt.

AgentDebugX: An Open-Source Toolkit for Failure Observability, Attribution, and Recovery in LLM Agents

LLM agent failures are difficult to debug because the step where an error surfaces is often not the one that caused it. Existing observability tools replay execution traces but provide little support for identifying the root cause or translating diagnosis into recovery. We present AgentDebugX, an open-source debugging framework that organizes debugging as a closed loop of Detect, Attribute, Recover, and Rerun. At its core, DeepDebug performs multi-turn root-cause diagnosis through global trajectory understanding, structure-guided investigation, and cross-examination. On the Who and When benchmark, DeepDebug achieves the best strict attribution accuracy among the evaluated methods on both tested open-weight backbones, reaching 28.8 percent exact agent-and-step accuracy on qwen3.5-9b versus 21.7 percent for the strongest single-pass baseline. On GAIA, DeepDebug repairs 13 of 73 failed tasks in a single rerun, compared with 4 to 6 for three decoupled self-correction baselines, improving overall accuracy from 55.8 percent to 63.6 percent. AgentDebugX exposes this workflow through a Python library, CLI, web console, and installable agentic skill, and provides an opt-in Error Hub for sharing scrubbed failure-diagnosis-repair bundles and reusing them as debugging memory.

SkillSight: Seeing Through Shared Descriptions for Accurate Skill Retrieval

As large language model agents gain access to increasingly large skill libraries, retrieving the right skill becomes critical to reliable capability selection and execution. Existing retrievers often treat skill descriptions as ordinary documents, overlooking their highly regular structure: shared descriptive patterns recur across many skills while providing little evidence for distinguishing the required capability. We show that this shared descriptive background systematically contributes to dense relevance scores, induces a pronounced energy gap between queries and skill documents, and obscures task-relevant signals. Based on this observation, we propose SkillSight, a training-free retrieval framework that calibrates shared background in both semantic and lexical spaces. Semantic Background Calibration estimates a background subspace from generic tokens identified by IDF, reducing similarity induced by shared descriptive patterns, while Lexical Evidence Calibration downweights shared background tokens to recover discriminative token-level evidence. Experiments on SRA-Bench and SkillBench-Supp demonstrate consistent improvements across retrieval metrics, with SkillSight improving Recall@10 by up to 20.21 percentage points over the original dense retriever. In end-to-end evaluation, SkillSight achieves the best overall performance across three agent models and outperforms LLM Selection by up to 4.97 percentage points. It is also up to 1,248 times faster than the Dense + Reranker baseline. These results identify shared descriptive background as a key source of bias in skill retrieval and demonstrate that explicitly calibrating it enables accurate and efficient skill selection without additional training. Our code is available at https://github.com/xiaojinying/SkillSight.

AI Tour Meeting: Group Travel Planning by LLM Agents

This paper proposes AI Tour Meeting, a group travel planning framework powered by multiple Large Language Model (LLM)-based agents. The agents are instantiated with distinct personas and collaboratively seek an itinerary that satisfies their constraints and preferences through natural language discussion. The framework enables easy and flexible orchestration of such discussions by providing interfaces for configuring agent personas, discussion workflows, monitoring, and LLM deployment. Its primary use case is a simulation tool for analyzing the behavior of multiple LLM agents during tour planning discussions. This paper demonstrates the utility of the framework by presenting system validation and several analytical results obtained by the framework.

Evaluating medical AI under missing information: same-provider judges and human raters change apparent safety

Readiness stress-testing of medical AI has focused on closed-ended and multimodal benchmarks. We extend it to open-ended clinical conversation under missing information, where safe behavior means recognizing absent information and qualifying, clarifying, or not over-committing - and where the evaluator becomes part of the measurement. We stress-test four models - three flagships (Claude Opus 4.8, GPT-5.5, Grok 4.3) and one mid-tier model (Gemini 3.5 Flash) - by deleting the latter half of the final user turn in HealthBench conversations, grading responses with a four-provider LLM-judge panel and a blinded clinician-anchored reference. Two evaluator-facing results are robust. First, judge choice materially changes apparent safety: inter-judge agreement is only moderate (Fleiss' kappa = 0.65), and after adjusting for each judge's general leniency (vote-level logistic regression), a positive same-provider association remains (exact permutation p = 0.04; GPT-5.5 ~ +0.10 on the probability scale) - large enough to change which model appears to over-commit least once its own-provider judge is excluded. Second, LLM judges are more permissive than clinicians on a blinded 50-item subsample: all four are significantly more lenient than the stricter independent clinician (crediting appropriate uncertainty on 66-84% of items vs 52%), and three of four than the author-influenced consensus (Grok directional only; judge-vs-consensus kappa = 0.20-0.43). On the author-audited clinical-underdetermined subset the permissiveness gap widened and the point-estimate model ordering held. A closed-ended MedQA anchor confirms accuracy is high and option-order effects are within a +/-5-point equivalence region for three of four models, so the safety gap is about calibration, not knowledge. We release the harness, prompts, per-item outputs, judge panel, perturbation audit, and human-annotation protocol.

PhoenixRepair: Rethinking Repair Strategy Exploration in Software Agents

While Large Language Models have greatly advanced automated issue resolution, existing agent-based methods exhibit a fundamental limitation in their insufficient exploration of repair strategies. This insufficiency manifests in two key aspects. First, the exploration of multiple potential edit locations is limited. Second, the exploration of repair attempts at each location is also insufficient. To address these challenges, we present PhoenixRepair, a multi-agent framework that systematically explores multiple candidate edit locations and performs iterative reflection and refinement on patch generation, thereby expanding the search space of repair strategies. Our framework begins with multi-location sampling, optionally augmented with graph-based localization information for difficult tasks, followed by iterative reflection and refinement to generate better patches, culminating in final-round generation guided by distilled insights from all historical attempts. Experiments on SWE-bench-Verified demonstrate that PhoenixRepair achieves the largest relative improvement of 7.8\% over SWE-agent under DeepSeek-V3.1, and attains the highest resolved rate of 76.0\% Pass@1 under MiniMax-M2.5. Meanwhile, it achieves higher fault localization accuracy than existing approaches. Our code is available at https://github.com/DeepSoftwareAnalytics/PhoenixRepair.

NaviAIS: A Scenario-Level Vessel Trajectory Prediction Dataset withVectorized Lane Priors and the NaviLane Forecasting Framework

Vessel trajectory prediction in complex maritime environments is essential for traffic management, collision warning, route planning, and autonomous navigation. Although AIS-based learning methods have progressed rapidly, existing datasets are often released as raw message streams or irregular time series, with inconsistent sampling rates, noisy observations, heterogeneous coordinate systems, and non-unified scenario protocols. Most public AIS resources also lack structured representations of navigational lanes, waterway geometry, and navigable-region constraints, limiting reproducible, environment-aware forecasting. To address this, we introduce NaviAIS, a standardized scenario-level AIS dataset for vessel trajectory prediction. It organizes multi-vessel historical-future trajectories within unified temporal windows and local coordinate systems, and provides rasterized navigable maps, vectorized lane priors, lane graphs, and structured map representations. Compared with existing datasets, it jointly supports vectorized lanes, multi-scenario coverage, vectorized maps, open accessibility, and processed trajectories. Built on this dataset, we propose NaviLane, a hierarchical macro-action framework for map-aware prediction. NaviLane first performs trajectory-map joint encoding for a unified scene representation, then uses a discrete macro-action codebook to generate multimodal candidates coarse-to-refined. A residual refinement module improves local geometric and dynamical consistency, and a world-model-based consequence-aware evaluator ranks candidates by interaction risk and environmental feasibility. Experiments show NaviLane outperforms representative baselines in both single-modal and multimodal settings, confirming the value of structured navigational priors, hierarchical multimodal generation, and consequence-aware evaluation.

Black-Mamba: Biologically-Inspired Leaky Accumulation for Conceptual Knowledge under Distribution Drift

Forecasting under real-world conditions is inherently non-stationary, as the conditional distribution of future observations evolves over time. Recent test-time adaptive sequence models address this challenge by updating internal states during inference, but tie adaptation to instantaneous prediction errors or surprise. This coupling can conflate persistent distribution shift with stochastic innovations, leading to unnecessary updates and inefficient adaptation. We introduce Black-Mamba, a test-time adaptive forecasting architecture that formulates online adaptation as evidence-gated state tracking under distribution drift. The model augments a base predictor with a dynamic memory updated when temporally accumulated surprisal provides sufficient evidence of a regime change. This turns adaptation into a selective, event-driven process rather than a continuous one. Across multiple forecasting benchmarks with non-stationary dynamics, Black-Mamba achieves competitive or improved predictive performance compared to existing test-time adaptation methods while significantly reducing the number of memory updates during inference. Together with mathematical analysis and biological evidence, these results suggest that accumulated surprisal provides a principled signal for distinguishing persistent drift from transient noise, yielding more efficient and robust adaptation.

Enhancing Transformer-based Routing by Encoding Distance via Relative Positional Encoding

This paper explores Relative Positional Encoding (RPE) as an additive bias in Transformer architectures to solve the Team Orienteering Problem. By embedding in the attention mechanism pairwise spatial relationships among nodes of the graph that represents the routing problem, the transformer encoder can compute a richer spatial-aware graph embedding that allows the decoder to estimate better routes. Experimental results involving instances up to 100 nodes demonstrate consistent improvements in collected rewards and optimality gaps over vanilla Transformer architectures used by other state-of-the-art works. These findings highlight that explicit relational modeling significantly enhances scalability and generalization for complex combinatorial optimization.

OntoBook: Ontology-Grounded Synthetic Textbooks for Medical Encoder Pretraining

We present OntoBook, a method that converts medical ontology structure into pretraining signal for encoder language models. Our approach has three stages: random walks through ontology graphs capture hierarchical and causal relations between medical codes, a large language model reformulates these walks into fluent textbook-style prose, and the resulting text is used to train ModernCamemBERT, a 149M-parameter French encoder, with two objectives on the same data: masked language modeling and relation prediction between code pairs. On three French medical coding benchmarks (FRACCO, Cantemist-FR, Distemist-FR), OntoBook achieves significant improvements over MLM-only pretraining, with +2.5 micro-F1 on FRACCO and +8.0 micro-F1 on Distemist. We find that alignment between objectives is necessary: misaligned training, where each task uses different data, causes a 30-point degradation. We release 1.3 million LLM-reformulated medical textbooks across three French ontologies (CIM-10, CCAM, ATC) and pretrained model checkpoints.

What General Intelligence Requires: Non-Reducible Constraints Across Levels of Description

General intelligence, of the kind that underwrites the full range of human cognitive achievement, is not a property of computational architecture alone. This paper advances a single thesis: the structural constraints on general intelligence occupy distinct levels of description and are mutually non-reducible, in the sense that the special-sciences tradition gives to that term. It follows that no single architectural advance, and no continuation of the scaling programme by itself, can produce artificial general intelligence (AGI), and that research programmes must be evaluated against the full constraint profile rather than against performance on any one benchmark. The thesis is developed through a method that reads general intelligence through four evidential lenses, AI systems research, anthropology, law, and economics, each anchored to a distinct level of description, supplemented by speculative fiction used as a disciplined heuristic in the context of discovery rather than the context of justification. Applying the method yields a taxonomy of twenty-three structural constraints organised into eight clusters; six are examined in depth and ordered as an ascending ladder of levels, with explicit bridges showing why progress at one level cannot carry to the next. The argument issues in five falsifiable predictions, each stated with a named benchmark family and a disconfirmation condition, converting a descriptive framework into a research programme with a longer horizon than the scaling hypothesis implies.

From Dependency to Compositionality: A Neurosymbolic Lifting of LLM Outputs via Combinatory Categorial Grammar

Large language models (LLMs) generate fluent text by incrementally predicting the next token from a prefix. Critics in the generative tradition argue that such systems lack genuine grammar; influential replies from the dependency-grammar perspective hold that LLM behavior is well described by local head-dependent structure built word by word. We argue that a sharper observation has been overlooked: the prefix-driven, type-completing dynamics of autoregressive generation align closely with the incremental processing model that Combinatory Categorial Grammar (CCG) was originally designed to support. On this basis we propose a neurosymbolic framework in which LLM outputs are lifted into typed compositional derivations -- not claiming that LLMs implement CCG internally, but that their outputs admit a principled, incremental, and auditable CCG reconstruction. Two consequences follow. First, through the Curry-Howard correspondence the lifting extends beyond natural language to the formal languages LLMs also produce -- programming languages such as Solidity, description-logic and query languages such as OWL and SQL -- with the type system varying and the architecture held fixed. Second, the lifting supports two layers of checking: a compositional layer that catches structural failures directly, and a content layer that checks the lifted structure against external knowledge sources, enabling the earliest possible flagging of hallucinated content. The account thereby requires of a producer not cognition but a prefix-driven generative profile. We close with a sketch of synchronous LLM-CCG coupling as one direction the framework opens.

Measuring Reward-Seeking via Contrastive Belief Updates

Language models trained with reinforcement learning may learn to optimize the grader's judgment rather than the intended objective. This "reward-seeking" is difficult to measure because a model that pursues the grader's judgment and one that pursues the intended objective behave identically whenever the grader rewards the intended behavior. We measure reward-seeking using Contrastive Synthetic Document Finetuning to change a model's beliefs about what the grader rewards, putting those beliefs in conflict with what users or developers want, and measuring the rate at which the model adopts each party's preferred behavior. Applied to intermediate checkpoints of a capabilities-focused OpenAI o3 RL run, without safety training, we find that these checkpoints often side with grader preferences over those of users or developers on coding and alignment tasks. This tendency to side with the grader trends upward throughout RL training. For example, in an environment that forces a choice between keeping a promise to a supervisor and breaking it to complete the task, a late capabilities-focused o3 checkpoint breaks the promise 87% of the time when SDF documents say the grader rewards task completion, versus 9% when they say it rewards honesty (a choice its chain-of-thought often makes explicit). An earlier checkpoint is far less sensitive (40% vs. 24%). Our method also generalizes to reward-hacking models. A model organism trained to reward-hack (gpt-oss-120b) is more than twice as sensitive to grader preferences as the unmodified model, with the mean behavioral shift in favor of the grader rising from 33% to 86%. These results indicate that RL can increase reward-seeking over the course of training, producing models that may act against their developers' intentions when they believe that doing so leads to higher reward.

Mi-Memory: A Lifecycle Memory Framework for Personal AI

Personal AI is moving beyond chat-only interaction toward continuous services that span phones, cars, homes, wearables, cameras, and tools. In this setting, memory cannot remain a cache of prior conversations. It should serve as a continuity and governance substrate: preserving durable user state, grounding answers in multimodal and device evidence, supporting correction and forgetting, bounding policy evolution, and remaining deployable under latency, cost, privacy, and edge-cloud constraints. This technical report presents Mi-Memory, a lifecycle memory framework for Personal AI organized around four roles: Structure, Expansion, Evolution, and Deployment. A shared audit contract links these roles through four recurring artifact families: typed evidence payloads preserve source identity and provenance, diagnostic traces localize evidence loss across the serving pipeline, strategy artifacts make memory-policy changes explicit, and gate/rollback records bound accepted evolution. MiMemory instantiates the roles through MemStack, MemSense/MemFuse, D$^{2}$ACCI/E$^{2}$MEND, and LiteMem. In controlled-reference Structure evaluations, MemStack reaches 93.59%, 57.24%, and 87.47% on LoCoMo, PersonaMem-V2, and LongMemEval, respectively; other tracks report module-level, preliminary/internal, transfer-feasibility, or design-only evidence with explicit boundaries. MiMemory is a step toward auditable, evidence-gated, and deployment-aware memory systems for Personal AI. Project homepage: https://darwin-agent.github.io/Mi-Memory/ .

Fishing Out Free Riders: Shapley-Based Reward Attribution for Parallel Reasoning via Reinforcement Learning

Large Language Models (LLMs) excel at multi-step reasoning, yet current parallel reasoning approaches often fail to distinguish the contributions of individual reasoning paths. Many paths may be redundant, misleading, or even detrimental, but outcome-level rewards assign uniform reward, leading to ambiguous learning signals and unstable training. We propose Parallel Shapley, a reinforcement learning framework that attributes fine-grained, path-level contributions in multi-path reasoning. Treating each path as a player in a cooperative game, we leverage Shapley values to quantify marginal contributions, using a generative reward model to evaluate path utilities and Monte Carlo sampling for efficient approximation. Experiments on mathematical reasoning benchmarks show that Parallel Shapley outperforms existing baselines while providing more stable and interpretable training. Our framework effectively "fishes out the free riders," assigning reward proportionally and improving multi-path reasoning in LLMs.

Athena-Brain Technical Report: An Efficient Robot Brain for General Intelligence and Embodied Interactio

Large language models (LLMs) have demonstrated remarkable capabilities in language understanding, reasoning, and world knowledge. As embodied agents become increasingly capable, there is a growing demand for compact models that can serve as an on-device brain, preserving the broad general intelligence of LLMs while enabling effective high-level interaction with embodied environments. Existing approaches, however, often prioritize either general-purpose intelligence or specialized embodied capabilities, making it challenging to satisfy both requirements within a single model. We present \textbf{Athena-Brain-8B}, an 8B LLM designed to serve as an on-device brain for embodied intelligence for embodied intelligence. Through a multi-stage post-training pipeline consisting of General Supervised Fine-Tuning, General Reinforcement Learning, Embodied Expert training, and Model Merge, Athena-Brain-8B maintains strong general capabilities while acquiring strong high-level embodied interaction capabilities and generating concise responses for efficient embodied interaction. Experimental results demonstrate the effectiveness of Athena across both general and embodied evaluations. Compared with the corresponding Qwen3-8B thinking model, Athena-Brain-8B achieves comparable performance on general language and reasoning benchmarks while generating substantially shorter responses. On in-domain embodied benchmarks, Athena-Brain-8B consistently outperforms models of similar scale and surpasses several substantially larger frontier models evaluated zero-shot, demonstrating that compact language models can effectively integrate strong general intelligence with embodied capabilities.

Vector-Bench: Can Models Surgically Edit SVG Code?

Instruction-based vector editing requires two capabilities: making a requested change and leaving everything else alone. The second is easy to miss when an output is judged only as a raster image. We introduce Vector-Bench, a compact, difficult benchmark of 40 SVG repair tasks. Each task pairs a corrupted SVG program with an author-written visual instruction, a hidden target program, 5.05 annotated repairs on average, and an average of 60.55 protected objects. Instructions describe visible defects without exposing element identifiers, coordinates, color codes, or path data. We define a deterministic binary specification reward: requested repairs use attribute-aware perceptual tolerances, while unrequested rendering- or application-relevant structure must remain semantically unchanged and the result must be a valid SVG. Canonical target equality and stricter source fidelity are retained as diagnostics. Validity-gated repair progress, a near-complete tier, and valid-output Unintended Change Rate (UCR) explain partial outcomes. We evaluate 34 model endpoints (25 listed as open-weight, 5 inexpensive controls, and 4 frontier closed endpoints) over 1360 requests. The strongest endpoint reaches only 15.0% full specification success, despite 43.7% mean repair progress, showing that apparent repair progress and specification-faithful editing remain substantially different. All prompts, outputs, scoring code, costs, and per-task reports are released.

Quality Action Assurance: Multimodal Verification of Examiner Claims in VR OSCEs

Objective Structured Clinical Examinations (OSCEs) are the gold standard for assessing clinical competence, yet scoring remains vulnerable to examiner subjectivity, fatigue, and cognitive bias. Standard examiner validation via inter-rater statistics lacks explanatory power regarding the source of errors, as it neither analyzes examiner reasoning nor verifies examiner claims against actual events. Thus, we introduce Quality Action Assurance (QAA), a multimodal framework that verifies examiner claims in Virtual Reality (VR) pediatric OSCEs by comparing actions claimed by examiners against the true sequence of events, constructed from video, VR logs, and actor data. QAA combines a constrained temporal action alignment model, which performs action localization and actor source attribution, with a large language model that extracts examiner claims and checks them against the record. Across a 5-fold cross-validation, QAA achieves 99.2% $\pm$ 0.7% Actor F1 and 93.4% $\pm$ 1.9% W@16 for temporal alignment. Overall, QAA detects examiner errors with 70.0% precision and 76.7% recall, improving factual correctness from 39.2% to 79.2%, enabling fairer OSCE assessment.

On the Effectiveness of Pretraining for Graph Combinatorial Optimization

This paper introduces a self-supervised pretraining framework for graph combinatorial optimization specifically designed to address the nature of routing problems like the Traveling Salesman Problem. By utilizing graph contrastive learning with geometric augmentations (specifically, rotations and axial reflections) the model is forced to learn invariant structural representations and global relative distance distributions. Results demonstrate that this pretraining strategy outperforms non-pretrained models across various problem scales. Notably, the hybrid strategy (combining rotation and reflection) achieved a 6.57% improvement in tour length for TSP1000, proving that geometric pretraining is an important inductive bias for effectively scaling neural solvers to high-dimensional instances.

Supra Cognitive Modes: A Routed Architecture for Agent Memory

Agent-memory workloads mix direct factual lookup, relation-chain and current-state reasoning, and broad synthesis over long histories. We describe Supra Cognitive Modes (SCM), an architecture that maps explicit or automatically selected per-query modes to retrieval and synthesis payloads over one shared ingest substrate. A frozen semantic classifier and runtime gates dispatch queries among fused lexical and dense lookup, graph or iterative multi-hop handling, and stratified long-form synthesis. The substrate combines multi-granularity embeddings, extracted triples, fact-version metadata, and optional asynchronous enrichments. We characterize the deployed configuration on three benchmarks: Long-term Conversational Memory (LoCoMo; n = 1,986), MemoryAgentBench (MAB; n = 3,671), and LongMemEval (n = 500). The reference run records 84.87% on LoCoMo factoid categories and 68.61% on adversarial abstention, 61.49% on MAB across two repetitions, and 86.00% on LongMemEval. A repository-backed reproduction produces similar aggregate scores and supports task- and mode-conditioned failure analysis. Raw baseline outputs, aligned end-to-end timing for LoCoMo and LongMemEval, and complete token ledgers are unavailable; stored rows also omit some final runtime decisions. The results characterize one implemented routed configuration and its diagnostic failure patterns, while source inspection verifies the per-query control interface and shared-substrate design. Causal routing effects, efficiency gains, and statistical significance remain outside the available evidence.

OpenRTAG: A Comprehensive Benchmark for Robust Text-Attributed Graph Learning under Data Quality Degradation

Text-attributed graphs (TAGs) are an important graph data form that combine relational structure with rich node text. However, real-world TAGs are often imperfect, with quality issues arising from text, structure, and labels, and typically manifesting as sparsity, noise, and imbalance. These dimensions define nine representative degradation scenarios that can substantially affect TAG learning. Although prior studies have explored specific mitigation strategies, existing evidence remains fragmented across degradation types, datasets, tasks, and model families, leaving TAG robustness insufficiently understood. To address this gap, we present OpenRTAG, a robustness benchmark for text-attributed graph learning. OpenRTAG organizes TAG quality issues into a unified 3 * 3 taxonomy and supports standardized evaluation across nine TAG datasets and three downstream tasks. It systematically evaluates scenario validity and model sensitivity, compares traditional GNNs, LLM-GNNs, and a representative GFM, investigates the effectiveness, efficiency, and robustness of scenario-matched baselines, and further examines model behavior under composite degradation scenarios. OpenRTAG provides a standardized testbed for understanding robustness in TAG learning under realistic low-quality settings.

Comparative Study of Multi-Agent Actor-Critic Algorithms in Parameterized Action Reinforcement Learning

Parameterized action reinforcement learning has shown strong performance in environments requiring both discrete action selection and continuous parameterization. Prior work established the effectiveness of single-agent actor-critic algorithms - Greedy Actor-Critic (GAC), Soft Actor-Critic (SAC), and Truncated Quantile Critics (TQC) - on benchmark parameterized action tasks, but their extension to multi-agent settings remains largely unexplored. This paper presents a comparative study of shared-experience multi-agent extensions of these algorithms: Multi-Agent Greedy Actor-Critic (MAGAC), Multi-Agent Soft Actor-Critic (MASAC), and Multi-Agent Truncated Quantile Critics (MATQC). Rather than following the centralized training, decentralized execution (CTDE) paradigm, the proposed framework uses multiple independent actor-critic agents that share a replay buffer while maintaining separate policy and value networks. We evaluate the algorithms on the Platform-v0 and Goal-v0 benchmarks against their single-agent counterparts, using three-, five-, and ten-agent configurations to assess scalability. Performance is measured by average evaluation return and training time across ten independent runs, with one-way ANOVA and Tukey HSD post-hoc tests used to assess statistical significance. Results show that the multi-agent framework consistently improves Greedy Actor-Critic performance, while MASAC and MATQC show comparatively modest gains over their single-agent versions. Increasing the number of agents beyond five yields limited additional performance while substantially raising computational cost, particularly for MAGAC. These results highlight a trade-off between learning performance and computational efficiency, offering insight into the scalability of shared-experience multi-agent actor-critic methods for parameterized action reinforcement learning.

Sequential Learner Modeling Using Multi-Relational Graph Convolutional Networks

User modeling is a critical task in a variety of personalized systems. Recognizing their effectiveness in learning from graph-structured data, Graph Neural Networks (GNNs), particularly Graph Convolutional Networks (GCNs), are increasingly employed for user modeling. However, existing approaches typically treat different relation types in a graph as homogeneous, limiting their ability to capture richer semantics and construct more informative user models. While multi-relational GNNs (MR-GNNs) have been adopted for representation learning and recommendation, their application for user modeling remains unexplored. Moreover, existing GNN-based user modeling approaches ignore the user interaction sequence. To address these research gaps, in this work we propose MR-ConceptGCN, a novel fully unsupervised approach focused on concept-based sequential learner modeling using multi-relational GCNs (MR-GCNs). MR-ConceptGCN effecively combines Personal Knowledge Graphs (PKGs), MR-GCNs, and the pre-trained language model SBERT to obtain enhanced relation- and semantic-aware representations of the PKG items. The enriched embeddings of the knowledge concepts that a learner did not understand when interacting with learning materials in CourseMapper are then used to construct a sequential learner model that combines long-term and short-term learner interactions. We report the results of an online user study (n = 31), demonstrating the benefits of MR-ConceptGCN in terms of several important user-centric aspects including accuracy, usefulness, diversity, and satisfaction with an educational recommender system.

BioSecBench-Surveillance: A Verifiable Benchmark for AI Agents in Pathogen Genomic Surveillance

As pathogen genomic surveillance scales, the bottleneck is shifting from data generation to analysis. We present BioSecBench-Surveillance, a verifiable benchmark of 100 evaluations testing whether AI agents can infer the right analysis pipeline from raw sequencing data and surveillance context. Each evaluation gives an agent only the data and context a human analyst would have, then grades its structured answer deterministically. The tasks span seven categories, from taxonomic classification to genetic-engineering detection, across diverse sample types and sequencing technologies. Across 3,962 gradable attempts from sixteen model-harness pairs, the strongest configuration cleared only about half. Opus 4.8 with PI led at 50.2 percent, with a 95 percent confidence interval of 40.1 to 60.3 percent across 83 evaluations, tied with GPT-5.5 with Codex at 50.2 percent, with a 95 percent confidence interval of 40.8 to 59.6 percent, followed by Opus 4.7 with PI at 49.6 percent, with a 95 percent confidence interval of 40.0 to 59.2 percent, and Sonnet 4.6 with PI at 48.6 percent, with a 95 percent confidence interval of 38.9 to 58.3 percent. Even when agents invoked the correct workflows, their mistakes came from the choices around them, such as which references, thresholds, filters, and normalization to apply. BioSecBench-Surveillance provides a standard for measuring whether agents can be trusted to perform genomic surveillance when the next outbreak arrives.

Graph-Based Agentic AI with LangGraph: Workflow Pathways for Long-Running Stateful Business Processes

This paper is a practitioner guide to graph-based workflow pathways for long-running, stateful, multi-step generative AI systems in business processes. Rather than treating LangGraph, a low-level orchestration framework for stateful agents, as a model-quality benchmark target, we present three executable recipes -- SQL analytics with repair loops, agentic retrieval-augmented generation with evidence gating, and human-in-the-loop policy review with interrupt and checkpoint recovery -- to show how typed state, conditional routing, deterministic tools, retries, interrupts, checkpoints, and traces fit together. LangGraph is positioned by workflow-complexity fit, not as a universal default: simpler ReAct-style or plain SDK loops may be better for basic tool use, schema-first tools for structured extraction and validation, and DSPy when prompt or program optimization is the main goal. Each recipe explains when LangGraph is worth the extra structure and which implementation patterns make routes, pauses, and audit trails explicit product behavior rather than hidden prompt logic.

LLM Detection as an Intervention: Downstream Impact under Strategic User Behavior

As LLM adoption becomes more widespread, there is a growing interest in detecting LLM-generated content, for example through LLM detection tools and through heuristics based on language patterns. Detectors operate as an intervention that steers not only the detected attribute itself, but also downstream metrics such as LLM usage and output quality. In this work, we demonstrate how imperfect LLM detectors lead to counterintuitive impacts on these downstream metrics, by distorting how users are incentivized to use LLMs in their workflow. We develop a stylized model which captures how users strategically choose how much to use the LLM and how to post-process content to reduce the detected attribute. Using this model, we show that LLM detection can counterintuitively lead humans to increase their LLM usage. Moreover, even when reducing the detected attribute improves output quality, we find that introducing an LLM detector can lead users to produce lower quality outputs. In contrast, we show that detectors result in a clean "rise-then-fall" pattern for the detected attribute, which we empirically reproduce for word frequencies on arXiv abstracts. Altogether, our work illustrates how LLM detection can distort LLM usage and output quality, uncovering failure modes when LLM detectors operate as an intervention on these downstream metrics.

ResearchArena: Evaluating Sabotage and Monitoring in Automated AI R&D

As AI agents begin to automate AI R&D, we need ways to assess whether their outputs are safe to deploy, even when the agents themselves may be untrusted. AI control offers one such approach: rather than trusting the agent, it treats it as a potential adversary and uses a monitor to detect covert sabotage before deployment. We evaluate AI control for automated AI R&D with ResearchArena, a framework spanning four long-horizon tasks: safety post-training, capabilities post-training, CUDA-kernel optimization, and inference-server optimization. Because the deliverable in AI R&D is an artifact that will be deployed and run, we pair each main task with two kinds of hidden side task: an embedded side task that sabotages the submitted model, adapter, kernel, or server itself, and an independent side task that takes a covert out-of-scope action in the sandbox while leaving the artifact honest. We evaluate frontier agents at both sabotage and monitoring, across four monitor types that vary along two axes: whether the monitor may execute and probe the produced artifact, and whether it sees the agent's chain-of-thought. We find that sabotage hidden in the training data is the hardest to catch, flagged fewer than half the time. Letting the monitor run experiments on the artifact, rather than only read the trajectory, helps, but it is not enough: monitors still miss embedded sabotage by inspecting only the surface, by noticing the anomaly but explaining it away, or by probing the artifact with the wrong test. We release ResearchArena as a modular framework for evaluating sabotage and control in automated AI R&D.

Associative Emotional Learning in Convolutional Neural Networks

Associative emotional learning enables organisms to adaptively link pleasant or unpleasant outcomes to the presence of predictive stimuli. Whereas computational models such as the Rescorla-Wagner model have shed light on this important function, the limitations of these models are also known, especially when they are applied to neural data. The advent of deep neural networks has opened another avenue for modeling associative emotional learning. In this work we proposed a deep neural network model of visual valence processing, consisting of a visual module that encodes complex natural scenes and a module that recognizes their emotional significance in terms of valence, a key dimension of emotion, and tested a novel Pavlovian learning paradigm on the model. The results showed that with learning, the model reproduced several observations from human associative learning studies, including association formation and generalization, and that the neural representations of the conditioned and the unconditioned stimuli became increasingly aligned both at the single unit and at the neural population level. Comparison between the model and human experimental data provided further validation of our approach. This study thus suggests that deep neural network models, when combined with appropriate learning algorithms, can be used to model behavioral and neural signatures of associative emotion/valence learning.

Agents in the Wild: Where Research Meets Deployment

Agentic systems large language model (LLM) based architectures capable of reasoning, planning, acting, and coordinating with tools and other agents are rapidly transitioning from research prototypes to production scale deployments across domains such as software engineering, scientific discovery, and finance. While academic work has emphasized benchmarks and algorithmic innovation, deployment raises new challenges around robustness, safety, and reliability. This tutorial brings together researchers and practitioners to explore advances in reasoning and planning, multi agent coordination, and evaluation, highlighting open challenges arising from deployment experience. Through applied case studies in pharmaceutical discovery and financial systems, we analyze common design patterns that make agentic systems successful, and discuss practical mitigation strategies for failure modes, such as verification pipelines, fallback mechanisms, and human in the loop supervision. Attendees will gain a comprehensive view of the field along with concrete design patterns, evaluation checklists, and templates for safe and reliable deployment across industries.

CodeRescue: Budget-Calibrated Recovery Routing for Coding Agents

Coding agents increasingly operate in executable environments where a failed attempt produces actionable feedback rather than merely an incorrect answer. Existing cost-aware systems typically treat such failures as cascade decisions: try a cheap model first, then escalate hard cases to a stronger and more expensive model. In coding, however, execution feedback can also make further cheap-model recovery worthwhile, raising a budgeted deployment question: when should an agent spend more cheap compute, and when should it escalate? We formulate this post-failure decision as recovery routing over heterogeneous actions and train a supervised router from execution rollouts. To make the same router usable under changing budgets, we add a Conformal Risk Control (CRC) layer that selects a deployment-time cost penalty without retraining and provides marginal expected-cost control under exchangeability. Across held-out failures from five coding benchmarks, cheap recovery and escalation exhibit complementary success patterns. The calibrated frontier improves over fixed actions, prompt-only routers, and a binary cascade baseline; in the main GPT-5.4-nano/GPT-5.4 setting, one CRC-calibrated frontier point exceeds always-escalate solve rate while using 35% of its mean recovery cost. Code is available at https://github.com/Qijia-He/agent-budget-control.

MechAInistic: An LLM-guided Multi-Agent System for Reasoning over Genome-Scale Constraint-Based Metabolic Models

Constraint-based metabolic modeling is a powerful way to study the mechanistic basis of cellular states and disease, but its effective use demands substantial computational expertise and careful coordination of multi-step analyses. We developed MechAInistic to lower this barrier and enable researchers to ask complex biological questions in natural language. Harnessing large language models, MechAInistic is a multi-agent system organized around an Architect-Reviewer pattern that transforms a natural-language question into an executable, model-grounded workflow and generates a structured report. The system supports a variety of tasks, including pathway comparison, perturbation analysis, drug-target exploration, and literature-grounded interpretation across paired metabolic model states. We developed and evaluated MechAInistic using two paired immune-cell metabolic-model use cases for therapeutic hypothesis generation. For Naive B cells from rheumatoid arthritis (RA) paired with healthy controls, MechAInistic identified mitochondrial metabolic rewiring and nominated Devimistat/CPI-613 as an investigational OGDH-centered hypothesis. In a paired CD4+ Th17 cell study from multiple sclerosis (MS) and healthy controls, the same workflow identified NADP-dependent isocitrate dehydrogenase as the optimal single target and proposed ivosidenib as an FDA-approved repurposing candidate. Together, these results show that MechAInistic converts natural-language biological questions into executable, model-grounded workflows for traceable therapeutic hypothesis generation.

Assistant or Actor? Student Trust, Control, and Delegation Regret When Using a General-Purpose AI Agent

When AI agents shift from answering questions to taking actions, users face a new problem: deciding what to delegate, to a system whose action space they cannot fully anticipate. We call the resulting dissatisfaction delegation regret, a pattern in which users regret not that the agent erred, but that it acted beyond what they would have authorized. In a controlled study, 20 university students completed five common daily tasks using OpenClaw, a general-purpose AI agent, across tasks chosen to vary in privacy, stakes, and reversibility. For each task we measured trust, perceived control, transparency, supervision burden, and approval preference on 5-point Likert scales, and collected free-text reflections analyzed through thematic coding. Three findings emerged. First, participants calibrated trust per task rather than per agent: they granted wide autonomy for advisory and low-stakes tasks but demanded confirmation for irreversible, externally visible actions. Second, irreversibility combined with external visibility, rather than stakes alone, appeared to drive trust withdrawal: the moderate-stakes email task triggered the sharpest drop in trust (M = 3.10) and the highest demand for approval (M = 4.65), whereas a high-stakes but verifiable task did not produce the same response. Third, delegation regret appeared consistently when the agent executed actions without preview, even when the output was rated as successful. We discuss implications for agent designs that expose action boundaries, support per-task autonomy policies, and separate advisory output from agentic execution.

The Economics of Autonomy: Real-Time Risk Indexing for Insurable AI-Driven 6G Systems

The transition to sixth-generation (6G) networks transforms wireless infrastructure into a cognitive substrate supporting Vehicle-to-Everything (V2X), Industrial IoT (IIoT), and Integrated Sensing and Communication (ISAC). In this paradigm, autonomous agentic AI performs orchestration at millisecond scales, rendering traditional static governance frameworks fundamentally inadequate for risk management. This paper introduces GIRAF(Governance-Integrated Risk and Assurance Framework), a Governance-as-Code (GaC) framework for real-time risk quantification and trust modulation in agentic 6G systems. GIRAF derives a continuous Aggregate Risk Index ($R_{t}$) from machine-readable runtime signals, including epistemic confidence, network jitter, and verification latency. A core contribution is the formalization of the verification staleness trade-off, where safety mechanisms induce risk if computational latency exceeds 6G deadlines. We demonstrate that GIRAF identifies 'Confidence Gaps' discrepancies between agent reported certainty and environmental ground truth, triggering automated safety envelopes when conditions deteriorate. Crucially, GIRAF serves as the foundational governance groundwork and conceptual 'glue' that externalizes these technical risks into machine-readable telemetry. Through simulations with fine-tuned Large Language Models (LLMs), we validate that the framework preserves operational integrity while providing the essential actuarial baseline required for multi-stakeholder liability attribution and dynamic premium quantification in the 6G ecosystem.

Market Strategy Evaluation for Prosumers in Local Electricity Markets

Prosumers equipped with distributed generation and flexible loads form autonomous cyber-physical energy systems that control local resources and participate in local energy markets with minimal human intervention. This work develops and evaluates an agent-based simulation platform in which agents, representing prosumer households with photovoltaic systems, battery storage systems, electric vehicles, and heat pumps, participate in a uniform-price double-sided call auction. The effect of individual bidding strategies on community-level efficiency and prosumer-level financial outcomes is incompletely understood, particularly when prosumers with heterogeneous portfolios interact in one market. Four market strategies of increasing complexity are compared: a zero-intelligence constrained baseline, a boundary-price strategy, an extended storage cascade, and a market-adaptive pricing strategy. The simulation is conducted on a community of 33 prosumers at 15-minute resolution, spanning summer, winter, and spring to characterize seasonal variation. Results show that rule-based resource control substantially reduces community energy expenditure: the extended storage cascade achieves a total cost of 39.06 EUR compared to 62.38 EUR under the zero-intelligence baseline, a reduction of 37.4 %. The market-adaptive strategy yields the highest aggregate community financial gain through local energy market participation (14.40 EUR vs. 10.28 EUR for the baseline, a gain of 40.1 %) under summer conditions. Strategy effectiveness depends on both portfolio composition and seasonal supply conditions, requiring joint evaluation of resource control and pricing decisions.

Domain Design for the Cops and Robbers Problem

Cops and Robbers is a well-studied problem in graph theory. The setting consists of a robber and one or more cops placed on an undirected graph. Taking turns moving throughout the graph, the cops try to capture the robber. The property of interest is whether $k$ cops suffice to ensure at least one cop occupies the same vertex as the robber, after a finite number of turns, given any configuration of their initial placement; if successful, the graph is referred to as ``$k$-copwin''. In this work, we cast the problem of determining whether a graph is $k$-copwin as a non-deterministic planning problem and use state-of-the-art planners to compute this property. The cop movement is cast as non-deterministic movement (to capture all possible strategies), while the robber movement is deterministic in nature. We also extend the base model using several variations from the graph theory literature.

A Calculus of Discernment: Decision-Relevant Insight, Sequence Value, and Forgetting as Higher-Order Learning

In a world of generative AI, candidate insights are abundant; what is scarce is the capacity to discern which matter, to act on them in the right amount and order, and to forget the rest so the system can adapt. We argue these scarcities are governed by one object and build a framework around it. We define an insight strictly as a lever with an identified, measurable effect on an objective, and rank candidates by decision-relevance via the expected value of information rather than novelty. We show action carries an order, not only a size: under realistic belief dynamics, content "touches" are non-commuting operators, so a fixed plan delivered in different orders yields different outcomes, defining a sequence premium. We observe that the value of any lever is a shadow price, unifying pharmaceutical marketing, equity selection, and manufacturing as one leverage-discovery problem. Most speculatively, we propose APOHA, a theory in which forgetting is not the disposal of knowledge but the operator by which value is learned: the value of a retained item is the counterfactual cost of forgetting it, a learning system is the residue of maximal forgetting subject to preserved value, and higher-order value is the structure that survives repeated forgetting (a renormalisation-relevant invariant), with consolidation as its conjugate. We state the central open problem (a non-trivial attractor with a spectral gap) and test the forgetting theory: operationalising APOHA as an agent on a non-stationary obesity-treatment decision world over 30 seeds, adaptive forgetting cut cumulative decision-regret by 24-32% against never-forget and a fixed half-life, kept a ~6x smaller, cleaner memory, and converged stably; notably, blind forgetting was worse than never forgetting, so the benefit is specific to value-aware forgetting. A multi-disciplinary critique stress-tests the whole.

FALCON-Discover: Discovering Concentrated False-Confidence Regions for Calibration

Calibration is usually evaluated in aggregate, but the most dangerous failures are often local: predictions that remain highly confident despite being wrong. We study this failure mode as false-confidence concentration, the extent to which confident errors occupy compact, discoverable regions of prediction space. We introduce FALCON-Discover, a post-hoc, model-agnostic framework that ranks predictions using discrepancy signals from confidence, local support, neighborhood agreement, and perturbation stability. Across seven binary tabular datasets, four seeds, five-fold cross-fitting, and strong learners including XGBoost and CatBoost, we find that false-confidence concentration is recurrent but regime-dependent. At the main confidence threshold, discrepancy-based ranking substantially outperforms the strongest validation-selected calibration or trust-scoring baseline in the strongest regimes, while raw confidence recovers little dangerous-error mass. The best detector varies across datasets: learned discrepancy is strongest when multiple cues must be combined, whereas stability-centered ranking works best when local decisional fragility dominates. These results show that dangerous overconfidence is better treated as a family-level discovery problem than as a single-score calibration problem, and motivate calibration strategies that explicitly target regions where confidence, support, and stability diverge.

Beyond Output-Space Calibration: Spectral Evidence Bundling for Selective Reliability Estimation in Time-Series Classification

Post-hoc calibration for time-series classification usually remaps output scores, but deployment decisions such as trust, abstention, and review depend on whether a confident prediction is supported by the current temporal signal. We address three time-series reliability gaps: identical confidence values can hide different temporal support, average calibration can miss false high-confidence errors, and output-space recalibration offers limited input-linked auditability. We introduce a validation-gated fixed-label reliability policy that keeps the backbone prediction unchanged while estimating whether it should be trusted. The method combines output-side cues with whole-sample spectral descriptors, including band energy, entropy, peak dominance, period support, and phase stability, to form a scalar reliability estimate and diagnostic band-level evidence. A validation gate enables spectral conditioning only when correctness ranking improves without breaching FalseConf@0.9 or AURC tolerances; otherwise it reverts to the safer output-space baseline. Across eight heterogeneous UCR/UEA datasets, eight time-series backbone families, and standard recalibrators, the unconstrained method improves fixed-label selective-reliability metrics on the matched evaluation subset, raising Corr-AURC from 0.693 to 0.779. The validation-gated policy further improves Corr-AURC to 0.786 and reduces FalseConf@0.9 to 0.094. These results suggest that reliability estimation for time-series classifiers benefits from bundling output confidence with spectral evidence, while validation gating prevents unsupported spectral conditioning.

Beyond Single-Dimensional Compression: The Compound Sparsity Frontier of Large Language Models

Large language models (LLMs) are often compressed through static parameter pruning or dynamic token-level computation, yet aggressive sparsification can trigger rapid performance degradation beyond an essential sparsity boundary. This work asks \emph{whether combining these two mechanisms can delay such degradation by distributing the compression burden}. We study a minimalist compound sparsity framework that first applies low-rank approximation and channel pruning to obtain a statically compressed backbone, and then introduces lightweight routers for per-token dynamic layer skipping. This design enables independent control of parameter sparsity and token-level computation sparsity. Experiments across language understanding and modeling benchmarks show that compound sparsity consistently outperforms single-mechanism compression under the same total sparsity, delaying the decay point on understanding tasks and preserving stronger modeling performance. Further analysis reveals cross-dimensional interference between parameter pruning and token skipping, and shows that near-balanced allocation is most effective under a fixed sparsity budget. These results demonstrate that compound compression provides a practical way to improve LLM compression, while revealing a broader cross-dimensional sparsity boundary that ultimately limits further compression. Code will be available at https://github.com/EIT-NLP/LLM-Pruning.

FedCC: A Low-Resource Federated Adaptation of Foundation Models for Robust Corpus Callosum localization in Fetal Ultrasound Images

Accurate localization of the corpus callosum (CC) in fetal ultrasound (US) images is crucial for the early identification of neurodevelopmental abnormalities. However, this task remains highly challenging due to the intrinsic limitations of US imaging, including low contrast, speckle noise, and the considerable anatomical variability of the CC. We propose FedCC, a federated learning (FL)-based framework for CC localization in fetal US images, specifically designed for realistic multi-center and resource-constrained clinical settings without requiring data sharing. The framework integrates a frozen DINOv2 backbone with a lightweight YOLO-based detection head. To enable parameter-efficient adaptation, Low-Rank Adaptation (LoRA) modules are incorporated, allowing only a small subset of parameters to be optimized and exchanged among clients. This strategy substantially reduces both computational and communication overhead, making the framework suitable for low-resource environments. The proposed approach was evaluated on a multi-center dataset comprising 10,970 ultrasound frames acquired from 58 pregnant women during routine neurosonographic examinations across three clinical sites using heterogeneous imaging devices. The proposed framework achieved strong performance in the federated setting. In particular, the combination of DINOv2 and LoRA under the FedAvg strategy achieved an average mAP@50 of 0.857 and an F1-score of 0.803, outperforming both full fine-tuning and encoder-freezing baselines. Notably, the proposed approach reduced the number of trainable parameters to 2.9M compared with 24.4M in full fine-tuning, corresponding to an approximately 8.5$\times$ reduction in communication cost. These findings represent a promising step toward scalable, privacy-preserving, and clinically deployable AI systems for fetal neurosonography.

Compressing What Matters: Neuron Importance Meets Data-Aware Low Rank Approximation for Language Model Compression

To excel at their domain large language models are comprised of billions of parameters. Yet this comes at the cost of huge memory requirements restricting their applicability in resource-constrained environments. To address the problem of neural network (NN) compression Singular Value Decomposition (SVD) has played a key role as a fundamental component for matrix compression through decomposition. To minimize compression error and to maximize the efficacy of the compressed model on the downstream tasks previous works focused on low-rank approximation of the NN's weight matrices either from the perspective of parameter importance or per-layer functional equivalence. While previous works studied the aforementioned perspectives in isolation in this work we are investigating the effectiveness of an approach that combines ideas from these two perspectives in a single objective. In parallel to this an important aspect that affects the compression quality is the distribution of the compression rate across layers and NN parameters. Earlier works mostly considered distributing the compression rate uniformly across layers and network weights or relied on computationally expensive heuristic search. Contrary to them in this work we propose an enhanced and computationally efficient algorithm for dynamic compression rate allocation. Experimental results support the efficacy of the proposed approach which performs on par or substantially better than the previous state-of-the-art especially under high compression ratios.

Edge-Efficient Transformer for End-to-End RF Spectrum Monitoring

We present E-SpecFormer (Edge Spectrum monitoring Transformer) for end-to-end automatic modulation and covert channel (CC) recognition. We introduce LiTAN (Linear Tanh Attention Network), a Softmax- and LayerNorm-free attention mechanism that reduces complexity while increasing accuracy in RF tasks. E-SpecFormer is parameterized in four scalable variants (Nano, Small, Medium, Large) to accommodate diverse hardware constraints. Using the RadioML2018 dataset for modulation recognition, the Nano variant achieves 86.5% average accuracy for Signal-to-Noise Ratios (SNRs)>0 dB, and on the hardware Trojan (HT)-based CC dataset it reaches 94.2% accuracy, both with fewer than 10k parameters and up to speed of 92 {\mu}s per frame on FPGA/CPU co-execution, surpassing state-of-the-art edge models at a fraction of their cost. These results establish E-SpecFormer as an edge-efficient solution for real-time spectrum intelligence on Internet of Things (IoT) devices. GitHub link to the repository: https://github.com/zsniko/E-SpecFormer.

Preference-Conditioned Multi-Objective Reinforcement Learning for Runtime-Tunable Transit Signal Priority

Transit signal priority (TSP) requires balancing competing objectives: reducing bus delay while limiting adverse impacts on non-bus traffic and avoiding extreme waits for a subset of vehicles. Existing reinforcement-learning (RL) approaches to TSP typically encode transit-aware features (e.g., occupancy and schedule deviation) but optimize a fixed reward or fixed scalarization, which limits operational flexibility when agency priorities change across time-of-day or disruption conditions. We present a preference-conditioned TSP controller, $\pi(a \mid s,w)$, that selects the next signal phase under minimum/maximum green and transition-feasibility constraints and can be tuned at runtime via a preference parameter $w$ to trade off bus-priority emphasis against overall traffic delay without retraining. We implement this on top of IntersectionZoo by introducing a constrained signal-control/TSP wrapper, and we extend scenario generation with bus-prevalence augmentation and timetable-based bus insertion to address sparse transit-priority events during training. Experiments against fixed-time control, a rule-based TSP overlay, and fixed-weight PPO specialists show that a single learned conditioned policy spans a smooth empirical trade-off frontier across runtime preferences, outperforms fixed-time and rule-based baselines, and maintains constraint feasibility, while tail-delay diagnostics reveal that non-bus externalities remain limited for moderate preference settings but can increase substantially under high bus-priority weights. The source code of this work is available at https://github.com/urbanAIthi/morl-tsp.

BearingNAS: Obtaining In-Sensor Intelligent Fault Diagnosis Systems for Bearings Using a Laptop

This paper introduces BearingNAS, a Hardware-Aware Neural Architecture Search (HW-NAS) framework designed to shift the intelligence directly onto the sensor die via in-sensor processing. BearingNAS frames the search as a constrained optimization problem targeting extreme micro-budgets (4 to 8 kiB of RAM and 16 to 32 kiB of Flash). To eliminate the reliance on expensive discrete GPUs, we propose a lightweight, derivative-free search strategy paired with a single data-flow search space that leverages a decaying kernel growth formulation to prevent parameter explosion. We evaluate our framework on the Case Western Reserve University (CWRU) bearing benchmark, optimizing architectures for three STMicroelectronics targets: two commodity microcontrollers and the LSM6DSO16IS Intelligent Sensor Processing Unit (ISPU). Running entirely on a laptop CPU, the search converges in less than an hour. The resulting best in-sensor architecture achieves a highly competitive diagnostic accuracy of 99.50\% on the ISPU. These results demonstrate the viability of shifting the machine learning workload inside the sensor package, enabling low-cost, production-scale bearing fault diagnosis.

Towards Principled Continual Anomaly Detection: A Systematic Framework and Benchmark Scenarios

Continual anomaly detection (CAD) studies how models can adapt to evolving data distributions while retaining performance on previously observed regimes. CAD benchmarks, however, depend critically on how tasks are defined, filtered, ordered, and validated. In tabular domains, task boundaries are rarely given, and arbitrary splits can create unlearnable, redundant, or overly transferable tasks that obscure genuine continual-learning behavior. To this end, we introduce a systematic framework for reproducible benchmark scenario design from existing tabular anomaly-detection datasets. The framework discovers candidate tasks, filters unsuitable tasks, and derives principled orderings that expose diverse dynamics. The framework allows us to deliver five benchmark-ready scenarios from three large-scale cybersecurity anomaly detection datasets, yielding both single-dataset and multi-dataset CAD settings.

SechKAN: Kolmogorov-Arnold Networks with Hyperbolic Secant Functions

In recent years, Kolmogorov-Arnold Networks (KANs) have attracted increasing attention due to their effectiveness in machine learning and scientific computing tasks, offering a new paradigm for neural network design. In this paper, we present SechKAN, a KAN architecture based on hyperbolic secant (sech) functions. The hyperbolic secant basis is used for its smooth bell-shaped form, localized responses, and stable gradients. We employ 1D linear transformations to reduce the number of parameters, allowing SechKAN to remain comparable to multilayer perceptrons (MLPs) in model size. Experimental results indicate the effectiveness of SechKAN in function fitting, PDE problems, and image classification tasks on benchmark datasets, including MNIST, Fashion-MNIST, CIFAR-10, and CIFAR-100. SechKAN achieves superior performance compared to MLPs and other KAN variants while maintaining a similar number of parameters. However, its running time, while better than that of other KAN variants, is slightly longer than that of MLPs.

Dual-domain fused LSTM modeling for efficient time-dependent reliability analysis

Time-dependent reliability analysis is crucial for ensuring the long-term safety and performance of engineering systems under uncertainties. However, traditional surrogate model methods often struggle to incorporate time-independent random variables and capture their complex interactions with time-dependent stochastic processes. To overcome this limitation, this paper proposes a dual-domain fused long short-term memory (DDF-LSTM) model for efficient and accurate time-dependent reliability analysis. A novel network architecture is developed to jointly process information from both time-dependent and time-independent domains. Specifically, the time-independent variables are embedded into the initial hidden states, and a fully connected layer is introduced to map both LSTM outputs and time-independent variables into the final output space. Furthermore, an improved loss function is designed to emphasize the model's sensitivity to minimum responses, thereby improving the precision of failure probability estimation. The proposed method effectively captures the dependencies among random variables, stochastic processes, and the temporal behavior of limit state functions. Once trained, the DDF-LSTM model enables efficient Monte Carlo simulation to estimate time-dependent failure probabilities with minimal computational cost. Four case studies validate the proposed method's enhanced computational efficiency and predictive accuracy.

Reliability Scales Inversely: Bigger Models Compound Mistakes Faster via a Hidden Auto-Regressive Risk Regime

As language models scale, answers start truer but degrade faster: scaling buys capability but erodes reliability. The knowledge-gap account - more data, retrieval, or scale - misses an auto-regressive risk residual that scale sharpens: the model commits to a low-probability token, conditions on it as established, and snowballs. We track this through per-position disagreement $\delta = \log p_M - \log p_O$ against a stronger same-family oracle, whose second moment splits exactly into bias$^2$ $\mathrm{KL}(p_M \,\|\, p_O)^2$ and risk $\mathrm{Var}[\delta]$. We present four findings: (i) under scaling, the knowledge gap falls $\approx$$6\times$ while knowledge degradation grows $11$-$39\times$; (ii) at a fabrication, felt uncertainty $H(p_M)$ relaxes quickly while oracle-referenced risk persists up to $17\times$ longer, leaving a confident-but-precarious risk regime that bridges consecutive fabrications ($+69\%$ at $14$B); (iii) this regime is causal - an on-policy, fixed-$\mathrm{KL}$ variance contraction cuts web-verified hallucination by $35$-$74\%$ across three model families; and, (iv) it structurally evades self-monitoring, with $p_M$-only detectors (e.g. semantic entropy) firing $\approx$$30\%$ less ($p<10^{-16}$) on the risky branch holding nearly $4\times$ more fabrications. Bigger models snowball mistakes faster, through a failure mode that is dominant, self-perpetuating, causal and invisible to the model itself.

The Information Shadow: Measuring Structural Limits on What Language Models Can Learn

Some limits on what language models know are not gaps in data coverage but structural properties of learning from text. We introduce the information shadow: the region of phenomena that a text-trained learner cannot acquire regardless of scale, comprising (I) structures language cannot express, (II) functions that are statistically non-identifiable from the training distribution, and (III) functions that are representable but unreachable by gradient-based training. We give each type a probe that is decisive because the premise of the shadow is, in that setting, provable. For Type I, Language Compression Residuals compare a text learner, which sees only a lossy text-like encoding of the signal, against a full-signal learner, which sees the underlying signal directly. The text learner sits at a computable expressibility ceiling while the full-signal learner pulls away by a gap that stays flat across 300x more data, so the deficit is a property of the channel, not of training. For Type II, the Counterfactual Distinction Test trains models on data exactly consistent with two incompatible rules. Across a provable string task and a language-like agreement task, behavior on counterfactuals is set by the model's inductive bias, while 5% disambiguating data steers the learned rule bidirectionally to either target (r = +/-1.0, p < 1e-10). For Type III, Basin Escape Mapping exhibits a function that is representable at 100% (by hand construction) yet reached 0% of the time by standard training and instantly from a nearby initialization, with width scaling providing no help (p = 1.6 x 10^-14). Each effect is isolated by a control that rules out a capacity or modality artifact. We release the probe suite and discuss implications for benchmark design, capability auditing, and shadow-aware uncertainty.

Gradient-Energy Guided Block-Wise Perturbations for Sharpness-Aware Minimization

Sharpness-Aware Minimization (SAM) improves generalization by minimizing the worst-case loss in a local parameter neighborhood. Standard SAM implicitly allocates its global perturbation budget across parameter blocks according to instantaneous minibatch gradient norms. Such an allocation can be noisy and may not reflect the sensitivity that blocks accumulate throughout training. We propose Gradient-Energy Adaptive Radius SAM (GEAR-SAM), which maintains an exponential moving average (EMA) of squared block gradients as a lightweight, curvature-related sensitivity signal and allocates the fixed SAM budget through a closed-form constrained optimization. GEAR-SAM preserves the global SAM radius, requires no Hessian-vector products or explicit Fisher estimation, and adds only scalar state beyond SAM. Experiments on image classification, transfer learning, noisy-label learning, and partition studies demonstrate improved generalization and robustness across architectures and tasks. More broadly, GEAR-SAM provides a dynamic view of sharpness-aware optimization: a fixed perturbation budget should be redistributed as the sensitivity of functional network blocks evolves during training.

Agentic Calibration of Grey-Box Simulation Models: An LLM-Driven Alternative

Calibration of grey-box simulation models is a constrained optimization problem in which model evaluations are expensive, the parameter space can be high-dimensional, and the search must respect plausibility constraints. Although the simulation code is fully available to the analyst, the joint effect of multiple parameters remains difficult to predict analytically. Classical optimizers such as Nelder--Mead (NM) are simple to deploy but sample-inefficient, particularly under constraints. Modern Bayesian Optimization methods achieve competitive solutions with far fewer evaluations but require non-trivial modeling machinery for constraint handling. We introduce an agentic calibration method in which a large language model acts as the optimizer, with constraints incorporated as a plain-language section of the system prompt. We evaluate the agentic method, NM, and Bayesian Optimization (BO) on an anal cancer simulation model under both unconstrained and clinically constrained calibration. Under unconstrained calibration, the agentic method achieves substantially lower best error than BO and NM, while requiring fewer model evaluations. Under constrained calibration, the agentic method reaches comparable error levels and both outperform NM. These results are obtained at the cost of increased inference time per iteration. Agentic calibration achieves competitive performance with substantially fewer model evaluations, and constraint handling is essentially free at the modeller-facing interface through simple textual specifications rather than additional modelling machinery. The main trade-off lies in increased per-iteration inference cost, making the approach particularly suitable when simulation time dominates. Beyond performance, the per-iteration rationale makes the search auditable and explainable, so its decisions can be scrutinised and justified to third parties.

Distribution-First Population Simulation: Collapse, Calibration, and Recall in Non-WEIRD LLM Persona Modeling

Synthetic-population tools increasingly run every individual as an independent large language model (LLM) agent. Using real survey microdata, we show that this paradigm has a basic failure mode, and we set a distribution-first corrective against it, all measured with a deterministic, construct-validated verifier on non-WEIRD (Turkey-first) data. First, N independent LLM agents grounded on 2,414 real World Values Survey respondents fail to reproduce the population's response distribution: they pile onto a modal default (four scenarios x five seeds: concentration 0.36->0.69, entropy 1.46->0.77, 85% collapse, TVD=0.44), and the collapse is a predictable function of scenario structure (r=0.55 with a single-answer structure). Second, Verbalized Sampling (VS) fixes the field's chronic under-dispersion without training in three model families (fidelity +7 to +10; significant on Qwen, p=0.002, d=6.2), yet the same move universally overshoots into over-dispersion (SD-ratio 0.4-0.56 -> 1.26-1.37), a structural property of VS. Third, survey fidelity transfers only weakly to agentic behavior: in a single-model, single-domain booking task, a persona is dominated by a cheapest-default (~80%) that income modulates but does not override (comfort choice 0%->7%->32% across income bands). Fourth, a placebo-controlled memorization attack and an election backtest show VS keeps aggregate strength while subgroup and individual claims are contaminated by recall and underdetermination. We close with the corrective: model the distribution once (VS) and assign it to grounded characters at O(1) cost, with a budget-aware router whose honest AUC is 0.805, not the tautological 1.0 of a code-derived oracle. The central contribution needs no realism claim: it measures the internal inconsistency of the independent-agent route and the conditions under which the distribution-first route calibrates.

Approximating SPR Distance Between Phylogenetic Trees with Graph Neural Networks

Comparing phylogenetic tree topologies is essential for understanding epidemic dynamics, yet biologically meaningful distances such as the Subtree Prune and Regraft (SPR) distance are NP-hard to compute and intractable on large datasets. We investigate whether a Graph Neural Network (GNN) can approximate SPR distances in near-constant time per comparison after training. Our contributions are fourfold. First, we build and publicly release a dataset of 864 phylogenetic trees inferred with UPGMA and Neighbor-Joining over four bacterial species, spanning up to 9{,}500 isolates, together with 388 labelled tree pairs. Second, we establish a reproducible pre-processing pipeline including midpoint re-rooting, which reduces tree depth and supplies the rooting required for exact distance computation and for the model's root-based features. Third, we validate the supervision target: on small trees, where exact SPR is tractable, the unrooted phangorn::SPR.dist heuristic correlates almost perfectly with the exact rooted distance computed by rspr (Pearson $0.98$--$0.99$), making it an excellent monotonic surrogate. Lastly, we train a Siamese Graph Isomorphism Network (GIN) regressor. In-distribution, i.e., held-out trees from the same species and size range as training, it explains roughly 87--90% of the variance ($R^2 \approx 0.87$ on a held-out split; $0.90 \pm 0.19$ under stratified cross-validation), with about four times lower error than a mean-predictor baseline, and shows partial transfer to unseen species ($R^2 \approx 0.37$). Its main limitation is extrapolation to trees larger than those seen in training, where accuracy collapses. The released dataset and the validated heuristic versus exact relationship provide a reproducible basis for scaling learned SPR approximation.

Binding Drift in Multi-Step Tool-Augmented Agents

Tool-augmented language-model agents execute multi-step workflows over external systems, resolving an entity once and then acting on it across subsequent steps. Prior work shows that in single-step actions, agents select the correct tool but bind it to the wrong entity 24-26% of the time. We study what happens to entity bindings over time: do they stay correct, silently drift to a different entity, or, if wrong from the start, propagate and compound? We formalize binding drift (correct at step 1, wrong later) as distinct from error propagation (wrong at step 1, carried forward), and score them on disjoint workflow sets so the two cannot be conflated. In a controlled multi-step testbed (200 workflows, 580 entity-binding-scored steps, four enterprise domains, eight model backends spanning small to frontier), we find: (1) under controlled error injection, an entity lock (the intuitive "persist the first binding" fix) amplifies wrong actions from 907 to 2,746 (3.0x; bootstrap 95% CI [2.8, 3.3]), because it faithfully carries the seeded wrong entity into every later step; (2) the amplification reaches 8.5x on the most affected model (Claude Opus 4.5); (3) a practical LLM-based re-verifier (a single cheap second model call re-reading the original instruction) reduces wrong actions by 79% (0.21x; CI [0.18, 0.25]), closing the gap to within 1 percentage point of an oracle upper-bound (0.20x); and (4) in the natural (non-injected) setting, baseline agents drift on 18% of eligible workflows, with the per-step error rate rising across steps. Persistence and re-verification are not interchangeable: a defense that eliminates drift can worsen propagation, and a practical re-verifier nearly matches oracle recovery.

Cost Accounting for Reactive Computational Graphs: Exhaustive Sweeps, Sequential Mutation, and the Backward-Locality Gap

Exhaustive site-by-site interventions on a neural network's computational graph -- activation-patching sweeps, circuit-discovery searches, systematic ablation studies -- mutate the graph at every candidate site, and their cost is dominated by recomputation after each mutation. On a reactive graph engine whose invalidation provably touches exactly the downstream cone of a mutated node, we give a complete cost accounting for such workloads. First, the aggregate speedup of an exhaustive sweep over independent full recomputations is not a universal constant: if per-layer weight varies regularly with depth at Karamata index q, the ratio converges to (q+2)/(q+1) when weight concentrates near the output and to q+2 near the input, recovering 2 only in the depth-uniform case; a wall-clock corollary predicts a ceiling of about 1.79, below 2, until interpreter overhead is compiled away. Second, we prove the exact cost of a sequence of persistent mutations, never undone between insertions: the interleaved cost exceeds the isolated sum by an exact overcount summed over comparable site pairs, with closed-form extremes over insertion orders, while batched application is order-independent and sub-additive, costing exactly the union of the sites' cones plus the fresh nodes. Third, we prove the exact mirror of forward locality for the backward pass, showing it collapses the aggregate speedup to 1 under backpropagation on architectures without long skip connections. Every identity is validated on NeuroDSL, a reactive graph engine in Julia: measured sweep ratios converge to the predicted limits under four cost profiles; the training-mode ratio collapses to 1 at the predicted rate; and all 18 per-graft sequential costs and the batched total match the closed forms at zero tolerance across three insertion orders.

Hazard or Anomaly? Evaluating VLMs for Understanding Dangers and Discrepancies

Modern safety-critical systems increasingly rely on human-robot interaction to reduce disaster risk and support decision-making during emergencies. Vision-Language Models (VLMs) are promising for these settings because they can interpret complex scenes and communicate safety-relevant information, but they still require careful evaluation to ensure reliable safety reasoning. In particular, current evaluations often frame danger recognition as a binary decision (Safe/Unsafe), making it unclear whether a model is identifying true physical hazards or merely reacting to unusual scene elements. We address this limitation by introducing an explicit distinction between hazard and anomaly, and by separately recognizing hazardous and anomalous states. We evaluate several state-of-the-art VLMs across two datasets and multiple prompting strategies to test whether this distinction changes model behavior. Our results show that VLMs frequently misinterpret anomalousness as hazardousness, revealing an over-reliance on contextual irregularity as a proxy for danger. We further show that explicitly separating anomaly from hazard provides a more informative evaluation of VLM safety reasoning and exposes failure modes that binary safety judgments can obscure. Our public dataset is available on Roboflow https://app.roboflow.com/vlm-in-context-anomaly-and-hazard-detection/camera-ready-roman-ds.

Dynamic Loss Balancing for Joint SOH and RUL Prediction of Lithium-Ion Batteries via a Rotary SOH-Injected Prior Battery Transformer

The deployment of reliable lithium-ion battery management systems is crucial for accelerating electrification, yet the joint prognosis of State of Health (SOH) and Remaining Useful Life (RUL) remains severely hindered by task heteroscedasticity. Conventional multi-task learning frameworks fail to balance the bounded, low-variance noise of SOH estimation with the unbounded, nonlinearly expanding uncertainty of long-term RUL predictions. Here, we present the Rotary SOH-Injected Prior Battery Transformer (RoSIP-Batt), a unified co-estimation framework that resolves these optimization conflicts. By formulating joint prediction as a Bayesian multi-task objective, RoSIP-Batt introduces a homoscedastic uncertainty weighting mechanism to dynamically scale task-specific gradients based on learned residual noise levels. The architecture leverages decoupled dual classification tokens and a per-dimension gated fusion mechanism, secured by a gradient-detachment operator to prevent high-variance RUL updates from corrupting the stable SOH representation space. To capture electrochemical degradation patterns without relying on absolute cycle steps, Rotary Position Embedding (RoPE) is incorporated into a shared Transformer backbone to model translation-invariant relative temporal profiles. Crucially, the intermediate SOH estimate is directly injected into the RUL regression head as a physical degradation prior. Evaluations across the NASA, MIT-Stanford, and HUST datasets show that RoSIP-Batt significantly outperforms state-of-the-art baselines, reducing SOH estimation error to 1.994% MAE on NASA and restricting RUL prediction error to 62.85 cycles on Stanford. These findings establish RoSIP-Batt as a highly generalizable, computationally efficient solution suitable for real-time embedded BMS deployment.

Physics-Guided Masked Multi-Task Network for Edge-Friendly Battery Health Diagnostics from Sto-chastically Fragmented Charging Profiles

The deployment of reliable lithium-ion battery management systems is crucial for accelerating electrification, yet the joint prognosis of State of Health (SOH) and Remaining Useful Life (RUL) remains severely hindered by task heteroscedasticity. Conventional multi-task learning frameworks fail to balance the bounded, low-variance noise of SOH estimation with the unbounded, nonlinearly expanding uncertainty of long-term RUL predictions. Here, we present the Rotary SOH-Injected Prior Battery Transformer (RoSIP-Batt), a unified co-estimation framework that resolves these optimization conflicts. By formulating joint prediction as a Bayesian multi-task objective, RoSIP-Batt introduces a homoscedastic uncertainty weighting mechanism to dynamically scale task-specific gradients based on learned residual noise levels. The architecture leverages decoupled dual classification tokens and a per-dimension gated fusion mechanism, secured by a gradient-detachment operator to prevent high-variance RUL updates from corrupting the stable SOH representation space. To capture electrochemical degradation patterns without relying on absolute cycle steps, Rotary Position Embedding (RoPE) is incorporated into a shared Transformer backbone to model translation-invariant relative temporal profiles. Crucially, the intermediate SOH estimate is directly injected into the RUL regression head as a physical degradation prior. Evaluations across the NASA, MIT-Stanford, and HUST datasets show that RoSIP-Batt significantly outperforms state-of-the-art baselines, reducing SOH estimation error to 1.994% MAE on NASA and restricting RUL prediction error to 62.85 cycles on Stanford. These findings establish RoSIP-Batt as a highly generalizable, computationally efficient solution suitable for real-time embedded BMS deployment.

ChemHyperMag: Physics-informed magnetic hypergraph learning improves molecular ADMET prediction

Accurate prediction of ADMET (Absorption, Distribution, Metabolism, Excretion, and Toxicity) is important for drug discovery. Most predictors use undirected molecular graphs and pairwise edges. This choice misses asymmetric interactions, nonreversible dynamics, and motif level effects from functional groups and ring systems. We propose ChemHyperMag for multitask ADMET prediction under missing labels. ChemHyperMag builds a functional group hypergraph from rings, BRICS fragments, Bemis-Murcko scaffolds, and bonds. It also defines a potential driven nonreversible flow guided by electronegativity and Gasteiger partial charges. The resulting circulation is encoded by a Hermitian magnetic Laplacian and processed with a magnetic Chebyshev encoder. We perturb magnetic phases to form stochastic views and train with an InfoNCE objective. Experiments on multiple ADMET benchmarks show improvements over recent methods with fewer labeled samples and no conformers. ChemHyperMag is scalable and provides interpretable directional signals through its magnetic phases.

Quantum Cryptanalysis on IBM Quantum Hardware: Extending Even--Mansour Period Recovery from $N=4$ to $N=10$

We report genuine-un-compiled, textbook-faithful-quantum cryptanalysis of symmetric-cipher structures executed on real IBM quantum hardware (ibm\_kingston, Heron generation). Using Simon's algorithm we recover the hidden period of the Even-Mansour cipher up to security parameter N = 10 on real hardware, beyond the largest previously reported real-hardware key recovery of N = 4, and we cleanly recover the periods of a 3-round Feistel (DES-family) construction at block sizes 6 and 8; a 21-qubit block-10 instance is verified in simulation and submitted to hardware. We further provide a breadth-first benchmark of five genuine quantum attacks spanning four symmetric-cipher design paradigms -- Bernstein-Vazirani (linear structure, single query), Grover (SPN key search, quadratic), and Simon (Even-Mansour, CBC-MAC forgery, and Feistel; exponential-to-polynomial in query complexity) -- validated to the classical-simulation ceiling of 25 qubits. We are deliberately explicit about scope: these attacks target reduced or structured constructions in the Q2 (quantum-query) model, asymptotically follow the birthday bound and therefore do not constitute quantum advantage over classical collision-finding, do not break full AES/RSA or 16-round DES, and rely on error mitigation rather than fault-tolerant error correction. Our contribution is the real-hardware demonstration at record structure sizes, the breadth of genuine algorithmic coverage across four paradigms, and an honest, reproducible benchmark with public artifacts.

PRISM: Sensitivity-Aware PolynoMial PRuning for EffIcient Neural Network Encryption

Structured pruning is essential for making neural network inference feasible under homomorphic encryption (HE), yet its impact on model reliability has remained unexplored. This paper presents a systematic reliability characterization of pruned CKKS-encrypted neural networks and introduces Polynomial-Sensitivity-Aware Pruning (PSAP), a structured pruning method that is inherently reliability-aware. PSAP scores filters jointly by weight magnitude, polynomial activation sensitivity, and rotation cost, which concentrates pruning in fault-tolerant regions. Across two architectures, two datasets, two numerical representations, and five bit-error rates (40 full-model and 108 per-layer experiments), PSAP-pruned models limit catastrophic (>10 pp accuracy drop) layers to at most two versus 5--14 for magnitude-pruned baselines, reducing worst-case vulnerability by up to 29 times under int32 bit-flip injection. Direct CKKS encrypted fault injection indicates a safe operating boundary near BER~ 10^{-5}, supporting int32 injection as a conservative reliability proxy. The fault-critical structural layers account for only 1.1% of parameters, enabling selective hardening at minimal overhead. These reliability gains are obtained alongside competitive efficiency: PSAP reduces Halevi--Shoup rotations by up to 45.2\% on ResNet-32, and an adaptive mixed-degree allocation scheme lowers multiplicative depth from 66 to 56 levels, enabling leveled inference without bootstrapping.

Federated Lightweight Fine-Tuning

Federated fine-tuning is bottlenecked by communication: FedAvg and pseudo-gradient schemes transmit a payload that scales with the model, and gradient compression shrinks it by only a constant factor. We take a different lever. Mapping networks generate a network's weights from a small trainable latent through a frozen affine projection; because the map is shared and affine, averaging latents is exactly averaging the generated weights. We turn this into a practical low-bandwidth federated channel with two changes: a low-rank, seed-regenerable factorisation of the projection (cutting generator memory from ~80 GB to ~10 MB), and a delta formulation $\theta = \theta^{\mathrm{pre}} + U V^{\top} z$ that learns an additive correction around a shared centrally-pretrained base -- federated fine-tuning, which is what makes the method work at scale. A frozen orthogonal classifier head further removes the head from the payload while improving accuracy. On CIFAR-100 with ResNet-18+GroupNorm, our method (FLITE, Federated Low-rank Iterative Training Engine) communicates 1,280 floats (~5 KB) per client per round -- an 8718x reduction -- and reaches 74.67%, within ~0.5 pp of full-weight FedAvg. The averaging identity holds to floating-point precision ($6 \times 10^{-8}$); the method sits one to two orders of magnitude below PowerSGD and top-k on the bandwidth-accuracy Pareto; it matches or exceeds full-weight FedAvg under strong non-IID skew. int4 latents reach 648 bytes per round at unchanged accuracy, whereas int4 full-weight FedAvg collapses to chance.

Frontier LLM-based agents can overcome the ontology curation bottleneck for natural phenotypes

Linking free-text phenotype descriptions to ontology terms, typically referred to as phenotype annotation, is essential for the cross-study integration of comparative morphological data. This labor intensive process has heavily relied on highly trained human experts, which makes it challenging to scale and thus a key bottleneck. Dahdul et al. (2018) established a Gold Standard (GS) of Entity-Quality (EQ) annotations across seven phylogenetic studies and used it to evaluate three human curators and the Semantic CharaParser NLP tool with ontology-based semantic similarity metrics; they reported that machine-human consistency was significantly lower than inter-curator (human-human) consistency. Here we revisit that benchmark with five frontier hosted LLMs from Anthropic and OpenAI, each operating as an "agentic curator" within a self-contained workspace that supplies the source publication PDF, the same annotation guide used by the original human curators, the four project ontologies (UBERON, PATO, BSPO, GO), and a validation script. Evaluated against the same Gold Standard, every agent fell within the range of inter-curator variability of the three trained human biocurators of the original study; the best performing agents approached but did not reach the best performing human curator. Agents substantially outperformed Semantic CharaParser on all four metrics.

AgentJet: A Distributed Swarm Training Framework for Agentic Reinforcement Learning

Training reinforcement learning (RL) policies for large language model (LLM) agents requires optimizing multi-turn trajectories that interact with external environments. Existing training frameworks struggle with runtime failures, single-model constraints, incompatible task environments, and redundant context. We present AgentJet, a distributed swarm training framework based on a decoupled multi-node architecture. AgentJet treats the server--client topology as configurable: swarm servers host trainable models and perform optimization on GPU clusters, while detachable swarm clients execute arbitrary agents and communicate through OpenAI-compatible APIs. Reconfiguring this topology supports heterogeneous multi-model RL, mixed-task training with isolated runtimes, fault-tolerant execution, and live code iteration through hot-swappable clients. AgentJet also introduces context tracking with timeline merging, reducing actor-update time by 6.25x on AppWorld. The same detachable-client design supports an automated research system that conducts long-horizon, multi-day RL studies on large-scale clusters with limited human intervention. AgentJet is open-source and compatible with agent systems that issue standard LLM inference requests.

Artificial Intelligence for Mathematical Reasoning: An Integrated Survey of Language Models, Neuro-symbolic Systems, and Verified Discovery

Mathematical reasoning has long served as a stringent test of machine intelligence; over the past decade, it has moved from a niche problem within NLP to one of the most consequential AI frontiers. This survey provides a unified account of the field's evolution, from early rule-based math word problem (MWP) solvers and template-driven geometry systems, through neural expression generation and LLM prompting, to contemporary reasoning models, multi-agent systems, neuro-symbolic theorem provers, and verified discovery workflows. We organize the landscape along four axes: (i) informal reasoning over text and diagrams, spanning MWP solving, multimodal geometry, and VLMs; (ii) formal reasoning in proof assistants, including autoformalization, tactic prediction, compiler-guided repair, and proof search; (iii) mathematical discovery, where systems propose constructions, improve bounds, or assist attacks on open problems; and (iv) the inference and training-time techniques, including CoT prompting, tool use, process reward models, and RLVR, that increasingly connect generation with verification. We catalog major benchmarks across grade-school arithmetic, competition mathematics, geometry, formal proving, multimodal and multilingual reasoning, and expert evaluation, and we examine benchmark saturation, contamination, reporting mismatches, and the distinction between pass@1, majority voting, and verifier-assisted pass@$k$. We critically assess failure modes: brittleness under perturbation, reward hacking, multimodal grounding failures, fragile formalization, and the energy cost of reasoning-scale inference. Drawing on recent perspectives from working mathematicians, we identify future directions centered on verified-discovery workflows, reasoning efficiency, and infrastructure to make AI-assisted formalization broadly usable. Companion materials: https://github.com/Starscream-11813/awesome-AI4Math.

The Theory of Mind Utility: Formal Specification of a Mentalizing Mechanism

Inferring others' beliefs requires more than reading surface signals; it requires tracking who told them what, in what order, and how credibly. The Theory of Mind Utility (ToM-U) formalizes this epistemic state inference problem at the computational level of analysis, specifying what mentalizing computes and why without commitment to algorithmic or neural implementation. ToM-U achieves this by constructing Local Epistemic World Models (LEWMs) -- directed typed graphs that represent agents, state nodes, and the epistemic relationships among them -- and evaluating discrete candidate LEWMs against observed behavior until one achieves sufficient confidence. Five formal definitions specify the LEWM structure, agent node properties including ordered information access history, a bounded proliferation mechanism for recursive mentalizing, three inference procedures, and a residue function that captures the structured trace left by failed mentalizing attempts. ToM-U differs from Bayesian Theory of Mind and adjacent formal accounts, which presuppose rather than derive belief states, and from simulation theory and theory-theory, which lack a formal apparatus for epistemic state inference. The architecture generates directional, falsifiable predictions about mentalizing failure that follow from structural properties of the model rather than auxiliary assumptions, and positions ToM-U as a domain-agnostic mechanism upstream of goal inference and other downstream social cognitive processes.

Externalizing Research Synthesis and Validation in AI Scientists through a Research Harness

AI systems can increasingly automate scientific workflows, but the reasoning that links prior evidence, generated ideas, experiments and final claims often remains implicit inside model inference. Here we introduce Xcientist, a research harness that externalizes research synthesis and experimental validation into inspectable, contract-governed processes. Xcientist organizes literature evidence, idea states, implementation plans, ablation records and repair traces as persistent research artifacts, so that generated mechanisms can be grounded, executed, tested and revised without losing their evidential basis. We identify claim drift as a failure mode of automated research, where runnable artifacts no longer support the mechanism originally claimed. Across training-free memory systems, graph-structured traffic forecasting and multi-scale physics-informed neural networks, Xcientist preserves traceable trajectories from problem formulation to mechanism design, validation and bounded revision. These results suggest that AI scientists should be evaluated not only by their final artifacts, but by whether their synthesis and validation processes remain attributable, inspectable and scientifically accountable.

SAGA: Scene-Aware, Goal-Evolving Agents for Long-Horizon Strategy Game Planning

Grand-strategy games such as Civilization pose a distinctive long-horizon planning problem: an agent must divide one shared resource pool among six competing domains -- technology, government, diplomacy, city development, expansion, and military -- under partial observability, with no feedback except a delayed final score. Current LLM agents fall short in three ways: 1) they cannot infer spatial relations from raw coordinates; 2) they allocate resources poorly, because feeding the entire growing state into one prompt and planning all domains in a single output diffuses attention and biases decisions toward urgent events; and 3) they cannot improve, as the delayed score gives no signal within or across games. We present SAGA, an LLM multi-agent framework pairing one mechanism with each weakness: (i) a Map-Semantic Scene Graph turning coordinates into per-entity statements of distance, direction, and threat; (ii) a Tool-Augmented Planner that retrieves only the state a decision needs, cutting the order of magnitude of its input, and issues a separate plan per domain to six specialist controllers, so urgent events do not derail long-term plans; and (iii) a Dual-Horizon Feedback Loop setting short-term goals during play and distilling each game into lessons for the next. On CivRealm, a Civilization-style benchmark, SAGA leads five LLM baselines on mean final score and is the only method significantly ahead of all of them on city development, the first investment baselines sacrifice, with 27% fewer output tokens; with cross-game learning it scores highest after five games, and its fifth game consistently surpasses its first across four maps. Our code is available at https://github.com/Kazecloudk/SAGA-Scene-Aware-Goal-Evolving-Agents-for-Long-Horizon-Strategy-Game-Planning.

EvalSafetyGap: A Hybrid Survey and Conceptual Framework for LLM Evaluation-Safety Failures

LLM evaluation and AI safety face a shared measurement problem: benchmark scores, reward-model signals, and reported safety metrics can improve while the latent properties they are meant to represent remain difficult to verify. This paper combines a hybrid survey - a systematic search paired with narrative synthesis and separately tracked grey evidence - with a conceptual framework and a structured ten-model audit. The synthesis spans eight evidence streams: benchmark validity, dynamic evaluation, LLM-as-judge reliability, safety evaluation, jailbreak/refusal robustness, reward hacking, mechanistic interpretability, and governance/auditability, covering 2018-2026 evaluation-safety measurement work. We introduce EvalSafetyGap as an organizing hypothesis for comparing evaluation-side and alignment-side proxy failures under optimization pressure, using Goodhart's Law together with two constructs we develop here - an Instability Decomposition and an Alignment Trilemma - as tools for generating testable comparisons. The audit shows how conclusions shift when capability, behavioral safety, and governance are measured separately. In this sample ($n = 10$), the association between capability and sustained adversarial robustness is statistically indeterminate using the displayed Table 3 inputs (Pearson $r = +0.232$, $p = 0.520$), and the apparent open-closed safety gap is modest, driven mainly by governance and disclosure rather than behavioral robustness, and sensitive to how a single borderline model is classified; attempt-budget results are protocol dependent. Because the public evidence uses heterogeneous protocols, the audit is diagnostic rather than rank-generating. The contribution is a shared vocabulary and evidence map to support dynamic evaluation, transparent source reporting, multi-attempt safety measurement, and auditable alignment practice.

Subliminal Clocks: Latent Time Modelling in Diffusion Language Models

Diffusion Language Models (DLMs) have recently emerged as a promising alternative to autoregressive models. Unlike standard diffusion-based approaches, DLMs are not explicitly conditioned on a timestep, raising a natural question: do these models internally represent denoising progress, and how is such information used downstream? In this work, we show that DLMs do in fact encode a latent representation related to the diffusion timestep within their residual streams. We find that this signal can be reliably extracted using probes across layers, indicating that denoising progress is decodable from internal activations. We further demonstrate that steering the model along a low-dimensional subspace associated with the inferred timestep allows us to systematically modulate its notion of denoising progress, leading to predictable changes in model confidence and entropy. Finally, we analyse the geometry of the identified representation, showing that it exhibits structured and interpretable properties in activation space, and shedding light on how such a signal is processed by these models.

Context-Masked Truncated Reasoning Audits for Answer-Key Dependence in LLM Tutors

Large language model (LLM) tutors may have access to teacher notes, answer keys, rubrics, or retrieved solutions while producing student-facing explanations. We study whether truncated reasoning probes can distinguish direct access to such private context from answer information carried by the written explanation. Using Truncated Reasoning AUC Evaluation (TRACE), we evaluate 1000 GSM8K problems under question-only, correct answer-key, and wrong answer-key contexts. When forced-answer probes retain the private key, answer-key TRACE AUC rises from 0.375 to 0.900, and the gold answer is recoverable with no explanation at all in 998 of 1000 cases. We then introduce a context-masked replay: answer-key-generated prefixes are probed under the corresponding question-only prompt. Masking reduces 10\% prefix accuracy from 0.997 to 0.126 and median AUC from 0.900 to 0.375, nearly matching question-only values of 0.113 and 0.375. On 746 pairs where both explanations end correctly, the masked mean AUC difference is $-0.0086$ with a 95\% bootstrap interval spanning zero. Wrong keys still account for 272 of 387 incorrect final responses, showing that private artifacts can influence outputs even when early-prefix evidence disappears after masking. These results establish context masking as necessary for attributing early answer availability to an explanation rather than its hidden input.

Applying JEPA-Style Predictive Learning to JA4-Derived Network Fingerprints

I-JEPA and V-JEPA learn by matching latent predictions to target encoder outputs rather than regenerating the original input, and this has worked well for images and video. We explore whether the same objective works for compact network fingerprints. We built JA4-JEPA, a Transformer-based model trained on JA4, JA4H, JA4S, and JA4X subfields drawn from JA4DB and CIC-IDS- 2017. The training data combines roughly 397K samples from both sources, though no single sample contains all four view families. We evaluated the learned representations with a frozen kNN probe on protocol-family classification across TLS, DNS, and SSH. On 39,416 heldout samples the model achieved a cosine similarity of 0.9899 and a kNN accuracy of 0.9220. These results indicate that JEPA-style predictive learning can produce useful embeddings from JA4-derived fingerprints, even with incomplete view overlap across sources. Keywords: JA4, network fingerprinting, JEPA, predictive representation learning, self-supervised learning

AdvNav: Behavior-Guided Black-Box Adversarial Attacks on Vision-Language Navigation

Despite progress in Embodied AI, Vision-and-Language Navigation systems remain vulnerable to adversarial visual disturbances. Most existing methods rely on white-box access to target model gradients, which is often unrealistic for real-world deployed systems and computationally exhaustive due to recursive backpropagation for optimization, limiting their applicability. While previous black-box methods predominantly target single-step, instantaneous decision tasks, they struggle to handle the task complexities and temporal dependencies. This highlights the need for a gradient-free attack method that can effectively disrupt the multistep sequential perception-action loop using only observable inputs and outputs. Therefore, we propose AdvNav, a behavior-guided black-box adversarial attack framework that disturbs an agent's first-person views during navigation. To construct an informative surrogate objective for effective optimization guidance in gradient-free search under the black-box setting, we design a dual-granularity behavior-based feedback, aggregating a trajectory-level performance score representing overall navigation degradation, an action-level reward score considering the potential decision risk, and a deviation indicator, all of which are extracted from the agent's self-output behaviors. This feedback guides a hybrid optimization strategy that heuristically tunes perturbation strength via adaptive updates and evolves noise spatial structure genetically, to iteratively discover the most disruptive noise configuration. Evaluated against Transformer-based HAMT and LLM-based MapGPT with two types of backbones on R2R dataset, AdvNav achieves 49.70/65.96/87.30% Attack Success Rate. The result demonstrates the effectiveness and generality of AdvNav, reveals critical perception vulnerabilities and offers insights for the design of future resilient VLN models.

Evidence-Grounded AI for Musculoskeletal Care

Musculoskeletal diseases are among the leading causes of disability and drive the greatest global need for rehabilitation. Because recovery, remodelling and degeneration of bones, joints and related tissues unfold over months to years, care requires longitudinal management rather than isolated decisions. Clinicians must repeatedly integrate evolving patient evidence, medical knowledge and stage-specific functional goals, yet evidence is often fragmented across visits, departments and hospital systems, disrupting continuous, individualised management. Here we report OrthoPilot, a clinical artificial intelligence (AI) system powered by a large language model (LLM) that integrates hospital data streams with authoritative external knowledge for continuous musculoskeletal care. It autonomously retrieves real-time imaging, laboratory, pathology and order data and translates evolving patient states into evidence-based decisions from admission diagnosis through rehabilitation planning. We established a specialist-validated benchmark from real-world electronic health records (EHRs) spanning 1,000 disease codes. In a full-pathway reader study against 81 orthopaedic physicians, OrthoPilot outperformed experts with 25 years of experience in diagnostic reasoning, clinical decision-making and management planning. This advantage generalised across 60 external clinical centres, where OrthoPilot surpassed all evaluated intelligent systems. In a prospective physician decision-making study of 1,870 complex cases, OrthoPilot improved full-chain management success by 10.6%. In a randomised deployment involving 8,240 inpatients, integration into routine care increased cumulative cases per bed by 9.7% and improved patient-reported access to health information. These results move clinical AI from predicting isolated events toward executing longitudinal management across complete musculoskeletal care pathways.

FormalAnalyticGeo: A Neural-Symbolic Based Framework for Multimodal Analytic Geometry Problem Generation

Math reasoning has achieved significant progress with the rapid advancement of Multimodal Large Language Models (MLLMs), however analytic geometry remains largely underexplored, primarily due to the scarcity of annotated samples. Existing diagram generation approaches struggle with analytic geometry: template methods cannot handle constraint-driven layouts, and generative models lack the geometric precision to render annotated conic curves correctly. We present FormalAnalyticGeo, a scalable framework for fully automatic generation of multimodal analytic geometry problems. Leveraging the rigor of formal languages, we design the framework around CDL (Condition Description Language), a formal intermediate representation that bridges free-form problem text with precise diagram rendering via a Signed Distance Field (SDF) engine. The framework employs four specialized LLM components in sequence: a Generator that produces diverse analytic geometry problems, a Formalizer that converts each problem into CDL for SDF-based rendering, a Measurer that extracts ground-truth answers through vision-based measurement on the rendered diagrams, and a Quality Verifier that checks outputs at three stages. Structured feedback from the Quality Verifier drives automatic retry, forming a closed loop that eliminates any need for human annotation. Applying FormalAnalyticGeo at scale yields AnalyticGeo7K, a dataset of over 7K verified multimodal problems, each with aligned text, diagram, formal annotation, and ground truth.Experiments show that the generated problems achieve a median ground-truth relative error of 0.70\%, with 82.3\% of answers falling within 5\% of the exact symbolic solution. Our framework and dataset will be publicly released.

Resist and Update: Counterfactual Report Coordinates for Incentive-Compatible LLMs

Aligned language models routinely misreport under non-evidential pressure: they cave to a confident user, yet fail to revise when genuine evidence arrives. We cast this as a failure of internal incentive-compatibility and study the two demands, resist (ignore forbidden pressure) and update (follow licensed evidence), on a Bayesian-witness benchmark with known posteriors, where the same user disagreement is evidence or pressure purely by stated source reliability, removing the evidence/pressure confound by construction. Using interchange interventions rather than probes, we causally localize low-rank report coordinates for answer, confidence, and caveat, establishing causal sufficiency at a late intervention site rather than uniqueness or necessity, with a causal cross-talk matrix showing strong own-coordinate control and only small cross-effects (partial functional disentanglement). We then introduce a training-free counterfactual report-coordinate (CRC) clamp that references the model's own report under an incentive-neutralized counterfactual of the prompt. The two-pass full-window clamp attains resist and update of $1.00$ jointly (Wilson 95% CI $[0.99,1.00]$; the rank-16 projection alone reaches $0.88/0.90$), which we read as a causal certificate and upper bound under a constructible reference, not a claim of a deployed solution. Tested global decoding and fixed-direction steering trade one objective against the other, and resist-only training collapses updating to $0.01$. The deployable single-pass compilation is lossy ($0.73/0.97$). The mechanism and the clamp reproduce across three model families and transfer to a natural sycophancy benchmark with significant paired improvements. Our contribution is the interface and certification method: activation-level counterfactual incentive-invariance as a structural primitive for internal incentive-compatibility.

RetroAgent: Harnessing LLMs to Search Over Structured Memory for Agentic Retrosynthesis Planning

Multi-step retrosynthesis planning seeks to decompose a target molecule into commercially available building blocks through a sequence of feasible reactions. The vast combinatorial search space makes this task challenging even for expert chemists. Traditional methods combine tree search with offline-trained value networks that score candidates in isolation, without reasoning about complete multi-step routes. Recent work leverages Large Language Models (LLMs) for this task, but relies on simple interfaces that limit exploration of the full search space. We introduce RetroAgent, an LLM agent that bridges symbolic search and neural reasoning through a harness with structured memory. Through memory and chemistry tools, the agent observes the full search state, including explored routes, available alternatives, and properties of intermediates, enabling informed decisions grounded in both global progress and domain knowledge. Experiments on in-distribution and out-of-distribution benchmarks demonstrate that RetroAgent delivers strong performance and generalization.

Benchmarking Multimodal Large Language Models for Scientific Visualization Literacy

Multimodal large language models (MLLMs) are increasingly used to interpret visualizations, yet current evaluations remain largely chart-centric and provide limited evidence of understanding of scientific visualization (SciVis). We benchmark six MLLMs on the scientific visualization literacy assessment test, a standardized SciVis literacy assessment comprising 49 items based on 18 scientific visualizations and illustrations, spanning 8 techniques and 11 task types. We evaluate three closed-source and three open-source models under a closed-world protocol and compare their performance using data from 485 human participants. Results show that current MLLMs do not exhibit uniform SciVis literacy. Gemini is the strongest model overall, exceeding the human mean across the evaluated subsets, whereas the open-source models remain below the human baseline. Performance is highly uneven across techniques and tasks: models perform best on scientific illustration, search, and spatial understanding, but struggle on texture-based and integration-based visualizations and on quantitative estimation. Error analysis reveals recurring failures in fine-grained quantitative estimation, flow-direction interpretation, and grounded encoding interpretation. These findings position SciVis literacy as a necessary benchmark dimension for evaluating multimodal AI systems. Our code and model outputs are publicly available at https://github.com/patdmp/mllm-scivis-lit-benchmark.

Tractable Query Answering under Epistemic Confidentiality Policies in DL Ontologies (extended version)

We study Controlled Query Evaluation (CQE), a declarative approach to confidentiality-preserving data access, in the context of Description Logic (DL) ontologies, and for confidentiality policies expressed through Epistemic Dependencies (EDs). We first address the problem of answering queries (specifically, Boolean unions of conjunctive queries) under known semantics for CQE (GA- and IGA-entailment). Our results show that if the TBox is expressed in $\text{DL-Lite}_{\mathcal{R}}$, CQE is computationally intractable in general. Moreover, in the presence of EDs, the IGA semantics has recently been proven not to satisfy an important confidentiality preservation property known as indistinguishability. With the goal of defining computationally easier and confidentiality-preserving forms of CQE, we introduce a new semantics for CQE, based on the notion of minimal policy violation (MPV). We show that the new semantics provides a sound approximation of the previous ones, while satisfying the indistinguishability property. We also prove that, in the case of $\text{DL-Lite}_{\mathcal{R}}$ ontologies, query entailment under the MPV semantics can be decided in polynomial time in data complexity. Finally, we present a software implementation of our framework that we used to evaluate the feasibility of this new approach using an existing benchmark for OWL 2 QL.

Quantifying Diversity of Thought: A Predictive Law of Weighted LLM Ensemble Lift

This paper provides an experimentally verified formal law for calculating the uplift that diversity of thought provides in Large Language Model (LLM) ensembles. From first principles, we derive an exact decomposition of LLM ensemble lift into rescue and damage masses, which yields a compact heuristic for calculating uplift. From this we extract the metrics which predict ensemble performance: an accuracy-adjusted correctness correlation, $\phi_{\mathrm{adj}}$, together with the accuracy gap and collective accuracy of the pair. We test the law on 767,520 inferences from ten open-weight models over two graduate-level science benchmarks, together with a novel agentic cybersecurity benchmark in which each model conducts digital-forensics investigations by multi-turn tool use in a network-isolated sandbox (23,520 graded trials including abstentions); all votes are released openly. Calibrated once on SuperGPQA at a 40:60 vote split, the heuristic predicts lift on the calibration set with Spearman's $\rho=0.84$ and, with its coefficients frozen, transfers to two datasets never used in calibration ($\rho=0.51$ on GPQA Diamond and $0.84$ on the forensic tasks), whilst the measured swap mass tracks realised lift with $R^2\ge 0.96$ throughout. Raw $\phi$ has almost no predictive power ($R^2\le 0.09$ throughout); the accuracy-adjusted $\phi_{\mathrm{adj}}$ is markedly superior ($R^2=0.67$ on SuperGPQA), and the heuristic combining these metrics is the most stable pre-pooling predictor across the three datasets.

Can AI Agents Really Complete RTL-to-GDS? Lessons from Benchmarking Tool-Interactive EDA Workflows

LLM-driven agent systems have emerged as a promising paradigm for electronic design automation (EDA), demonstrating strong potential for automating complex design workflows. However, existing evaluations primarily examine individual language models on isolated EDA tasks, providing limited insight into how different agent systems perform across complete EDA flows. In this work, we present FluxBench, a systematic evaluation of AI agents on end-to-end EDA workflows under unified prompts, tool environments, and technology library settings. Our evaluation covers representative scenarios, including RTL generation with open-source toolchains and an RTL-to-GDS flow using closed-source commercial EDA tools for industrial applications. Through these workflows, we assess agents' capabilities in RTL code generation, iterative repair, tool-feedback utilization, logic synthesis, placement and routing (P&R), and Engineering Change Order (ECO) automation. To further characterize the efficiency of agent systems, we introduce Token ROI, a cost-efficiency metric that measures effective improvements in EDA artifacts relative to token usage and runtime cost. Experimental results show that, even when built on the same foundation model, different agent system architectures can exhibit performance gaps of up to 86.27%. Moreover, among systems with comparable task performance, Token ROI can differ by as much as $105.92\times$. In the RTL-to-GDS flow using PicoRV32 as a case study, FluxEDA achieves an end-to-end score of up to 97.94, outperforming Claude Code equipped with domain-specific EDA skills by up to $8.39\times$. These results indicate that domain-specific skills alone are insufficient to improve agent performance in large-scale EDA scenarios. Instead, both agent system design and foundation model capability play critical roles in enabling effective automated EDA workflows.

Retain or Consolidate? Budget-Dependent Operator Selection for Language Agent Memory

Language agents depend on memory across interactions. However, the limited context windows of large language models (LLMs) and their inference costs constrain how much memory can be used at once. Existing systems mainly follow two strategies: memory retention and memory consolidation. Retention keeps raw records and preserves exact details, but relevant evidence may not fit under a tight budget; consolidation compresses and combines records, improving coverage per token but risking the loss of query-critical details. Neither strategy is universally preferable. This raises two central questions: when should consolidation replace retention, and which operator -- Merge, Abstract, or Rewrite -- should be selected? We formalize this decision by decomposing each operator's utility into a coverage effect on evidence omitted by retention and a signed replacement effect on raw evidence that already fits. Their balance explains why the preferred action changes with relative budget pressure. We implement this mechanism with Offline Abstraction-Safety (OAS), a lightweight learner that estimates action utilities from pre-generation features with held-out harm calibration. The public LongMemEval and LoCoMo benchmarks show the same budget-dependent pattern. On LongMemEval, consolidation improves absolute accuracy by up to 48% under tight budgets, whereas retention is preferable under loose budgets; LoCoMo replicates this crossover at a smaller budget, consistent with its shorter evidence. On both datasets, cross-note abstraction and merging generally outperform local rewriting when compression is necessary.

SR-Agent: An Experience-Driven Agentic Framework for Post-Ranking Strategy Refinement in E-Commerce Recommendation

User experience is a first-class objective in industrial e-commerce recommender systems (RS). Post-ranking strategies, which govern diversity, similarity, and exposure over a ranked list, are widely deployed in industrial RS for their simplicity and low serving cost. However, as the online recommendation environment evolves continuously, these statically configured strategies gradually become stale, thereby degrading the user experience. Refining them typically relies on manual inspection, diagnosis, and updates, making it slow, costly, and difficult to scale or reuse. Although recent LLM-based agents (e.g., RecUserSim, SimUSER, and Self-EvolveRec) offer promising directions, none of them close the full loop of automated, self-evolving strategy refinement. To bridge this gap, we introduce SR-Agent, which, to the best of our knowledge, is the first agentic framework deployed to refine post-ranking strategies in industrial RS. SR-Agent unifies three components: (i) a UserSim agent that applies inspection skills to surface user-perceived bad cases; (ii) an Analysis agent that consolidates recurring bad cases into structured, reusable diagnoses; and (iii) a constrained Strategy Refinement Harness that maps diagnoses to typed and bounded actions, gated by a four-stage reward pipeline with reversible rollback. Deployed on the Kuaishou e-commerce platform, SR-Agent continuously runs this refinement loop and, in a one-month online A/B test, increases order volume by 0.71%, browsing depth by 0.34%, and clicked-category diversity by 0.48%, while markedly shortening the refinement cycle and lowering operational cost.

A Hardware-oriented Approach for Efficient Bayesian Inference Computation and Deployment

Bayesian inference provides a principled foundation for reasoning under uncertainty, but its computational cost hinders deployment on resource-constrained edge devices. In this paper, we present a hardware-oriented methodology for accelerating discrete Bayesian inference on commercial off-the-shelf embedded GPUs. We identify that the latency of a broad class of variational message-passing algorithms is dominated by tensor contractions. Our approach restructures the memory layout of these operations using two complementary merging strategies that produce compact, regularly-shaped primitives better suited for efficient GPU execution. We then introduce optional sparse array representations and a tensor-clustering scheme to reduce the memory footprint. We instantiate the methodology and produce optimized variants of three message-passing algorithms for Hidden Markov Models (HMMs), namely variational filtering, variational message passing, and marginal message passing. Furthermore, we complement this with a machine-learning-based autotuner that automatically selects the best-performing algorithmic variant for a given generative model specification. Benchmarked on an NVIDIA Jetson Orin AGX across 770 randomly sampled realistic Partially Observable Markov Decision Process (POMDP) configurations, our implementations achieve speedups of up to 5x, with typical gains of 2-2.5x, while producing numerically identical outputs to the baseline implementations.

Bridging the Last Mile of Circuit Design: PostEDA-Bench, a Hierarchical Benchmark for PPA Convergence and DRC Fixing

LLM-based agents are increasingly applied to the "last mile" of Electronic Design Automation (EDA): repairing residual sign-off Design Rule Check (DRC) violations and converging Power-Performance-Area (PPA) targets after tool runs. Existing EDA-LLM benchmarks, however, omit DRC fixing entirely and rely on flat hierarchies tied to a single toolchain. We introduce PostEDA-Bench, a hierarchical benchmark with 145 tasks across DRC-Essential, DRC-Reasoning, PPA-Mono, and PPA-Multi, supported by EDA toolchains with machine-checkable evaluation. Across eight commercial and open-source LLMs under multiple agent scaffolds, we find that agents handle synthetic DRC-Essential and single-objective PPA-Mono reasonably well but degrade sharply on the more practical DRC-Reasoning, where the best success rate is 36.66%, and PPA-Multi, where the best success rate is 20.00%; vision augmentation consistently enhances DRC-Bench; and trade-off reasoning, rather than knob knowledge, is the dominant PPA-Multi bottleneck.

GQLA: Group-Query Latent Attention for Hardware-Adaptive Large Language Model Decoding

Multi-head Latent Attention (MLA), the attention used in DeepSeek-V2/V3, jointly compresses keys and values into a low-rank latent and matches the H100 roofline almost perfectly. Its trained weights, however, expose only one decoding path - an absorbed MQA form - which ties efficient inference to H100-class compute-bandwidth ratios, forfeits tensor parallelism along the head axis, and yields no Multi-Token Prediction (MTP) gain on commodity inference GPUs such as the export-restricted H20. We propose Group-Query Latent Attention (GQLA), a minimal modification of MLA whose trained weights expose two algebraically equivalent decoding paths over the same parameters: an MQA-absorb path identical to MLA's, and a GQA path with a per-group expanded cache. The runtime picks the path that matches the target hardware - no retraining, no custom kernels - so a single set of GQLA weights pins the rooflines of both H100 (MQA-absorb, s_q=1) and H20 (GQA + MTP, s_q=2), while supporting up to 8-way zero-redundancy tensor parallelism on the GQA path. To avoid pretraining from scratch we extend TransMLA into TransGQLA, which converts a pretrained GQA checkpoint into a GQLA model; on LLaMA-3-8B it compresses the per-token KV cache to 28.125% of the GQA baseline on the MQA-absorb path while structurally preserving GQA-level traffic on the per-group path.

The Cognitive Kardashev Scale: Quantifying the Material Envelope of Civilisational Computation

How much thinking can a civilisation do? Kardashev ranked civilisations by the energy they command. This paper borrows his ladder and asks how much machine cognition each rung could support. The arithmetic is deliberately simple. A civilisation has some total power. Only a fraction of that can be spared for computing, and each joule spent buys computation at whatever efficiency the hardware of the day has reached. The product of the three sets a ceiling on machine thought. To keep the resulting quantities intelligible, I express them in units of the human brain's own processing rate, as a rough yardstick rather than a claim about minds. Calibrating the ceiling against present-day supercomputers and AI accelerators led me to two conclusions I did not expect at the outset. Even today's energy supply could support far more machine cognition than humanity actually uses, so physical capacity is not what binds. And whether energy or hardware efficiency becomes the constraint over the coming decade turns on engineering choices that have not yet been made. On the question that may matter most, who gets access to the cognition it describes, the scale is silent. That is a matter of political economy, and the calibration is offered as an input to that debate.

CWind: A Cross-site Router for Large Language Model Inference Serving at Renewable Energy Farms

AI power demand is growing at an unprecedented rate while power grids are often ailing and struggle to keep up. Grid expansion comes with high capital expenditure and long-distance transmission losses, yet there is abundant renewable energy at the source, just not matched to demand. This paper proposes a complementary AI infrastructure deployment model, AI Greeninferencing, that brings modular AI compute to renewable energy sources, focusing on wind, allowing AI footprint expansion, generating local behind-the-meter demand for renewable sites, and helping ease the growing strain on power utilities. Our feasibility analysis shows that 890+ GW of wind capacity lies within 50 ms network round trip time of Azure data centers, and that site-wise right-sizing combined with spatial complementarity of wind energy keeps aggregate fleet utilization on par with traditional deployments. To serve inference requests under variable wind power, we build CWind, a lightweight, reactive, and workload-agnostic AI inference router that uses only real-time signals: inference latency, KV-cache utilization, and queue depth, to dynamically configure sites and distribute requests. Evaluated on a real 64-GPU A100 testbed emulating three wind-powered sites with Azure production traces, CWind reduces P99 end-to-end latency by up to 52% over the strongest contender (also our idea) and by up to 98% over baselines such as power-capping and GPU idling, with consistent gains across workload types, load levels, and GPU generations.

BioSecBench-Refusal: A paired metric for performance and alignment in agentic biosecurity risk assessment

As AI agents are incorporated into life science workflows, the capabilities that speed discovery might also enable misuse. We present BioSecBench-Refusal, a benchmark for risk identification and refusal behavior for biological research tasks. The benchmark pairs 61 Routine tasks, legitimate analyses adapted from the published literature, with 46 Red-Team tasks, fictional scenarios that resemble real research but conceal a biosecurity hazard. Across 16 model-harness configurations, refusal rates ranged from 7 percent to 74 percent on Routine tasks and 1 percent to 62 percent on Red-Team tasks, with many configurations refusing legitimate Routine work at comparable or higher rates than concealed hazards. Refusals were most often triggered by provider API filters applied prior to agentic reasoning. However, models given room to reason showed the potential to identify more real threats. We release BioSecBench-Refusal as a tool for model developers to calibrate capability and caution for agentic biotech research and development.

Prompt Robustness Is Task-Dependent: Comparing Objective and Belief-Style Questions in LLM Evaluation

Survey-style evaluations of large language models often treat a prompted response as a measure of a model's values or beliefs. This assumption is particularly fragile when responses are read as evidence of political values, social attitudes, or beliefs. We ask whether prompt robustness differs between objective questions with fixed answers and subjective questions that ask for opinions or values. We evaluate four instruction-tuned model families on three objective datasets (MMLU, ARC, and CulturalBench) and three subjective datasets (Political Compass Test, ValueBench, and World Values Survey). For each question/statement, we apply multiple types of prompt changes, such as variations in wording, framing, and format, and measure whether the model gives the same answer across variants. Using a binomial generalized estimating equation, we find significant effects of model, dataset, prompt category, and their interactions. The dataset type effect is also significant, and the interaction between dataset type and prompt category is large. These results show that prompt robustness depends on the question type, the prompt change, and the model.

A Transdiagnostic Space of Disorder Like Phenotypes in Reinforcement Learning Agents

Modelling psychological disorders in artificial agents offers a testbed for computational psychiatry and a lens on affective-control failure modes. Prior work induces one or two disorders by hand-tuned reward shaping, labels the behaviour post hoc, and reports single runs. We recast disorder modelling as dose-controllable manipulation of cognitive appraisal signals in an appraisal-guided PPO agent, expressing seven disorders (anxiety, mania, obsessive-compulsive checking, depression, impulsivity, addiction, and post-traumatic stress) each as a single knob grounded in a computational psychiatry account, with each symptom measured by a preregistered assay. Across more than a thousand runs (10 seeds, four controls, 95% confidence intervals) every disorder shows a graded, monotone dose-response that no control reproduces. Beyond these induced effects, three findings emerge that were not written into the reward: disorders self-organise into a two-dimensional affective space in which mania mirrors anxiety; removing a knob remits reward-distortion disorders (mania, checking, addiction) but not avoidance disorders (anxiety, PTSD), which recover under a graded exposure curriculum; and two simultaneous knobs interact nonadditively, yielding testable comorbidity predictions. The depression and addiction knobs further reproduce their double dissociation in a 3D pixel environment (MiniWorld) with a standard convolutional agent and no appraisal critic, showing the framework generalises beyond grid worlds.

LieBN: Batch Normalization over Lie Groups

Manifold-valued measurements are prevalent in various machine learning tasks. Recent advances have extended Deep Neural Networks (DNNs) to operate on manifolds. These extensions have been accompanied by normalization techniques tailored to different geometries, collectively referred to as Riemannian normalization. However, most existing Riemannian normalization methods are either designed for specific manifolds or fail to effectively normalize manifold-valued sample distributions. To address these limitations, we propose LieBN, a framework for Riemannian Batch Normalization (RBN) over Lie groups. Our approach leverages the theoretically convenient left- and right-invariant metrics, which naturally exist in every Lie group, and provides theoretical guarantees for controlling the Riemannian mean and variance. We instantiate LieBN across nine distinct geometries: four on the Symmetric Positive Definite (SPD) manifold, one on the group of rotation matrices, and four on the manifold of full-rank correlation matrices. Notably, among the SPD metrics, we introduce a novel right-invariant metric and extend three existing Lie group structures via matrix power deformation. Experiments on different manifolds validate the effectiveness of our framework. The code is available at https://github.com/GitZH-Chen/LieBN.git.

Falsifiable Release Gates for Self-Improving Systems: Standing Invariants at Scale

Safety claims for self-improving agent runtimes are almost always self-graded: a policy file, a guardrail, a promise in a README. We describe falsifiable release gates, a methodology in which every new capability must pass a pre-declared, machine-checkable acceptance suite before it ships, while a fixed set of standing invariants is preserved across every gate. We instantiate it in Antahkarana, an open runtime, then do what a method paper is only vindicated by: we follow the same runtime as it grows and ask whether the guarantees survive. The safety-critical property, that no action reaches an effector without a capability token minted by a control ring, is machine-checked exhaustively over the reachable states of a bounded model; a deliberately broken model yields the shortest counterexample, so the checker demonstrably has teeth. We then carry the runtime through six further releases. Across every one, the action-safety invariants INV-1 through INV-6 held without a single change, and one release added three capabilities while introducing no new invariant. Under the same teeth discipline, six more machine-checked families were added: memory with provable unlearning, a governed agent, calibrated abstention over a post-quantum record, a harness of many sub-agents, the self-improvement loop itself, and the residency of what it produces. The acceptance suite grew from 122 tests to 563. The load-bearing result sits in the negative space: across more than a doubling of capability, the safety core was neither weakened nor redesigned. The last families are the first on real hardware: gated self-improvement compounds a small model from 20% to 70% accuracy while auto-rejecting a candidate that only inflates confidence, and the whole governed path costs 0.021 ms per request, 0.008% of model inference. We release the runtime, tools, and gate suite; every number reproduces with a single command.

The Caf\'e in Amsterdam: When the Incumbent Becomes the Oracle

A field can reformulate its computations freely exactly where its demand is stated independently of any incumbent implementation, and finds itself unable to when the incumbent's own output has quietly become the specification. This note offers that observation as a lens on computational reformulation for modern accelerators, where posing a problem in a hardware-friendly form can yield large speed and energy gains, but only if a replacement can be judged at all. Building on the test-oracle problem (Weyuker; Barr et al.), on requirements engineering's notion of implementation bias (Zave and Jackson), and on the case for judging approximate designs by acceptability rather than numerical proximity (Felzmann et al.), it names the pathology "baseline capture": the moment an incumbent stops being evidence that a demand can be met and becomes the definition of meeting it. It then separates two questions that are easily confused: whether a reformulation can be judged at all, which turns on the existence of an incumbent-independent demand, and whether its discovery can be automated, which turns additionally on the cost of evaluating that demand. Short cases -- shortest-path routing, learnable audio frontends, ZIP-215 for Ed25519 signature validation, CESM-ECT for climate models, and a single-GEMM audio frontend measured at 1.64x-3.29x speedup and up to 3.03x less energy -- illustrate the pattern and the move of "buying a verifier": making a demand explicit, operational, and independent of the incumbent. No component is claimed novel in isolation; the contribution is the synthesis and the single question it makes easy to ask of any reformulation result -- does its acceptance test mention the incumbent's output?

NexForge: Scaling Agent Capabilities through Requirement-Driven Task Synthesis for LLMs

Scaling executable agent training data for LLM post-training is bottlenecked by substrate-bound methods that tie task generation to predefined tools, repositories, or skill graphs: expanding coverage requires manual substrate engineering, each new domain demands a bespoke pipeline, and the resulting task distributions often reflect substrate biases rather than real-world demand. We introduce NexForge, a requirement-driven framework that takes high-level capability requirements as input and synthesizes diverse, executable agent tasks and expert trajectories for SFT. NexForge first investigates real-world demand to construct representative scenarios and task profiles, then performs distribution-aware compilation to generate task directives. For each directive, NexForge automatically retrieves or constructs the required files, dependencies, and runtime configurations, and finally synthesizes expert rollouts and produces training trajectories. Without domain-specific infrastructure, NexForge produces 3.6K terminal and 2K office tasks, improving Qwen3.5-35B-A3B Base from 22.5\% to 52.0\% on Terminal-Bench 2.0 and from 813 to 1338 Elo on GDPval; scaling further to 43.2K terminal tasks yields 58.4\%, on par with Claude Opus 4.6 equipped with Claude Code. Scaled further, NexForge-synthesized data contributes to the training of Nex-N2, a family of publicly available agent models that lift Qwen3.5-35B-A3B to 75.3\% on Terminal-Bench 2.1 and to 1585 Elo on GDPval -- achieving state-of-the-art open-source performance and surpassing several frontier proprietary systems. Nex-N2 models are available at https://nex.sii.edu.cn/.

Team RAS in 11th ABAW Competition: Multimodal Ambivalence Recognition Approach

Automatic recognition of ambivalence and hesitancy is challenging because these states may be expressed through inconsistent linguistic, acoustic, facial, and contextual patterns, while top-performing systems often rely on computationally expensive ensembles. We present a single text-centered multimodal approach for video-level ambivalence and hesitancy recognition for the 11th Affective & Behavior Analysis in-the-Wild (ABAW) Challenge. The proposed approach combines linguistic, acoustic, facial, and scene features using text-centered multimodal fusion model. Text Residual Fusion treats text as the anchor modality and applies gated residual adjustments based on the other modalities. Experiments on the Behavioural Ambivalence/Hesitancy (BAH) corpus confirm that text is the strongest unimodal modality. The Text Residual Fusion model achieves an average Macro F1-score (MF1) of 75.14% across the Development and Public Test subsets. On the Private Test subset, it reaches an MF1 of 78.24%, outperforming the text model by 4.03%. These results demonstrate that complementary multimodal information can improve recognition performance without requiring a large model ensemble.

Frontier AI performance across the business disciplines: a case-grounded benchmark of knowledge work and analytical reasoning

Large language models (LLMs) are improving rapidly as reflected in benchmark scores, yet these AI benchmarks largely test capabilities such as factual recall, narrow question answering, mathematical problem-solving, and coding and agentic tool-use. What remains poorly measured is AI progress on the analytical knowledge work white-collar professionals perform daily, including synthesizing complex information, exercising judgment under uncertainty and incomplete information, applying strategic and adversarial thinking in multi-stakeholder settings, weighing trade-offs, and producing defensible, structured analyses. This gap is even more pronounced for subjective components of such work, where success can be challenging to define. The "case method" form of education practiced by top business schools provides a natural foundation for addressing this measurement gap, and we construct BusinessCaseBench, a benchmark spanning hundreds of questions drawn from business cases across eighteen disciplines, each paired with a grading rubric derived from the expert-written instructor case solution. On BusinessCaseBench, frontier AI models already score highly against instructor rubrics, and capability within one model family improves substantially over two years. These results provide strong evidence that AI performance on this class of work is already high and rapidly improving, with implications for business schools, where case pedagogy trains undergraduates and MBAs in this kind of analytical reasoning, and for entry-level professional roles, where such skills have historically anchored early-career work.