Empirical Study

Despite recent advances, LLM-based web agents still struggle with limited exploration, omission of critical steps, and sensitivity to task constraints. Prior work suggests that many of these failures stem from weaknesses in planning, yet the impact of alternative natural language plan representation remains unexplored. To address this, we introduce PlanAhead, a static planner-executor framework that evaluates the impact of plan representation in agent performance. We first automatically categorize WebArena tasks into 3 difficulty levels, enabling consistent difficulty grading without human annotation. Then we systematically evaluate 4 different plan representations on the tasks categorized as hard: sequential subgoals, narrative, pseudocode, and checklist; across different families of multimodal LLM powered agents (OpenAI, Alibaba, and Google). To account for stochastic variability, we introduce two novel evaluation metrics: Achievement Rate (AR) and Solved-Task Consistency (STC). Our results show that both, the plan formulation and the underlying LLM generating the plan, significantly influence web-agent robustness and task success.

Deceptive alignment, in which models maintain accurate internal representations while deliberately producing false outputs, remains a central challenge in AI safety. While strategic deception is the primary long-term concern, synthetic dishonesty - induced via direct optimization on incorrect answers - provides a controlled testbed for studying the representational basis of learned deception. We introduce a multi-model paradigm in which honest and deceptive variants of five transformer models (Pythia-1.4B, Gemma-2-2B/9B, Qwen2.5-7B, Llama-3.1-8B) are fine-tuned using LoRA on the same question distribution. Linear probes trained on mean-pooled hidden states detect synthetic dishonesty with near-perfect AUC (greater than or equal to 0.99) as early as layers 1-3 in four architectures, while Pythia-1.4B reaches a peak of 0.705. Logistic regression probes consistently match or outperform MLP probes, supporting the Linear Representation Hypothesis. Probes trained on TruthfulQA generalize with near-zero loss (Delta AUC approx. 0) to held-out MMLU subjects. Late-layer representations show strong robustness to Gaussian noise, with Gemma-2 models exhibiting exceptional stability. Mechanistic analysis of Fisher Discriminant Ratio, effective rank, centroid geometry, directional stability, cross-domain alignment, and calibration (ECE) reveals two regimes: representational collapse in Pythia/Llama/Qwen versus high-dimensional preservation in Gemma-2. Across all models, the dishonesty direction consolidates progressively in deeper layers, with optimal calibration (ECE less than 0.01 except Pythia) achievable in layers 1-4. These results demonstrate that robust, domain-invariant dishonesty representations can be rapidly entrenched via modest supervised fine-tuning, with implications for activation-based monitoring.

Large language models (LLMs) are increasingly used for the automatic evaluation of generated text, yet most prior work focuses on English. Despite the growing demand for multilingual evaluation, extending LLM-based evaluators to multilingual settings remains challenging, particularly for low-resource languages and scenarios where in-domain data is scarce. This work explores several strategies for developing multilingual LLMs-as-a-judge, considering whether in-domain data is available for fine-tuning or not. We systematically analyze English, Spanish, and Basque, representing high-, mid-, and low-resource languages, considering instruction translation, monolingual versus multilingual supervision, and model size. For evaluation, we extend two existing meta-evaluation datasets to Basque and Spanish. Our results reveal key trade-offs: When in-domain data is available, fine-tuned smaller models can achieve performance comparable to proprietary models, whereas zero-shot evaluation with larger models proves more effective in out-of-domain settings. We also observe that fine-tuning on out-of-domain data can adversely affect model performance. These findings provide practical guidance for building efficient, reliable multilingual evaluation pipelines. The data and code are publicly available at hitz-zentroa/mJudge.

The rapid growth of computer vision and increasingly complex image recognition tasks has exposed fundamental computational limitations of classical machine learning models, motivating the exploration of quantum computing as an emerging new paradigm. This paper presents a comprehensive benchmarking study of classical and quantum machine learning models for image recognition on the MNIST handwritten digit dataset, evaluating both traditional models, a Classical Support Vector Machine (CSVM) and a Quantum Support Vector Machine (QSVM), and deep neural network models, a Classical Convolutional Neural Network (CCNN) and a Quantum Convolutional Neural Network (QCNN), across four performance dimensions: classification accuracy, computational runtime, parameter count, and memory requirements. Experiments are conducted as functions of both feature dimensionality and sample size, and across CPU and GPU execution environments, providing a controlled, multidimensional comparison to address gaps in prior work. For the SVM-based models, QSVM consistently outperforms CSVM in accuracy, reaching $\sim$ 0.90 versus $\sim$ 0.85 at 1,000 samples, with a higher computational cost. A feature count of 10 qubits and a sample size in the range of 200 -- 500 emerge as practical operating points that balance accuracy and runtime. For the neural network models, CCNN and QCNN achieve comparable classification accuracy, both exceeding 0.96 at 64 features and 60,000 samples, yet QCNN offers substantially superior parameter and memory efficiency, requiring $\sim$ 94% fewer parameters and $\sim$ 75% less memory than CCNN at higher feature counts, while incurring higher runtime. Across both model families, quantum models consistently outperform classical models by greater margins in accuracy as feature dimensionality or sample size increases.

Recent advances in artificial intelligence (AI) and social media data have led to growing optimism about the ability to detect suicide risk at scale. However, the empirical foundations of this work remain unclear. This article provides a synthesis of current research on AI-based suicide detection in social media, drawing on a recent umbrella review of 22 systematic reviews covering studies up to 2022, alongside an ongoing literature review extending the analysis to more recent work. Across these sources, we identified 195 relevant studies, which are documented in a detailed supplementary dataset outlining their key characteristics and findings (see Supplementary Information). Analysis of these studies reveals consistent patterns, including rapid growth, concentration on a small number of platforms, reliance on textual and English-language data, and repeated use of similar datasets. Most importantly, the majority of studies rely on indirect labeling strategies that do not involve direct, individual-level validation of suicide risk. Instead, ground truth is typically inferred from observable features of online content, such as linguistic markers or community membership. As a result, the predictive task often shifts from identifying individuals at risk to classifying posts that contain suicidal or distress-related language, limiting the ability of current approaches to detect individuals who do not express such content explicitly online. These findings suggest that current advances in model performance should be interpreted with caution. Progress in this field is likely to depend less on improving model performance and more on ensuring that model predictions meaningfully correspond to suicide risk as it is experienced in real life.

Adversarial training (AT) remains one of the most reliable empirical defenses against adversarial attacks. Its robustness critically depends on how the underlying min-max objective is optimized. In practice, Stochastic Gradient Descent (SGD) optimizer remains the default optimization choice for AT, whereas adaptive optimizers often improve standard training but may yield inferior robustness. Recently, the Muon optimizer, which orthogonalizes matrix-valued updates via an approximate polar decomposition, has achieved notable success in large-scale training at a memory cost comparable to SGD. This raises a security-relevant question: \textit{can orthogonalized optimization improve AT under strong and heterogeneous threat models?} Focusing on this problem, we conduct a comprehensive theoretical and empirical study. Theoretically, we show that Muon imposes a spectral-norm stability ceiling on matrix updates, limiting uncontrolled spectral growth in the training dynamics without explicitly shrinking the learned weights. Empirically, across five architectures and three $\ell_p$ threat models ($\ell_\infty$, $\ell_1$, $\ell_2$) and their union, Muon is competitive with SGD on CNNs and substantially outperforms AdamW on both CNNs and ViTs. These results identify optimizer geometry as a security-relevant factor in adversarial training, while clarifying the empirical regimes in which orthogonalized updates are beneficial. Overall, our findings highlight optimizer design as a security-critical component of AT.

Scientific data annotation, such as tracking animals in video or proofreading neural reconstructions, remains bottlenecked by the "last mile" problem: even with strong automation, verification and correction consume substantial human effort. Standard approaches train models to directly predict annotations, discarding the rich supervision in how experts navigate, click, verify, and correct. We introduce a framework for studying behavioral cloning on scientific annotation: 9 synthetic tasks paired with synthetic annotations that simulate realistic human strategies including exploration, mistake correction, and strategic decision-making. Our experiments reveal several findings. First, skills emerge hierarchically: models learn GUI mechanics before task-critical decisions, and commit fewer mistakes than the training data while retaining the ability to correct errors when they occur. Second, scaling models on multi-task behavioral cloning shows that larger models are more data efficient within our scale range. Third, multi-task pretraining enables efficient fine-tuning to new tasks, while training from scratch fails entirely. Fourth, linear probes reveal that models internally represent latent variables of the annotation process such as task phase and data position; interestingly, we find a shared mistake representation that generalizes across different annotation tasks. Overall, our framework establishes systematic benchmarks and identifies key bottlenecks, providing a foundation for scaling behavioral cloning to real-world scientific data annotation.

Multi-turn Text-to-SQL is central to enterprise analytics yet remains predominantly evaluated in single-turn settings. We introduce EnterpriseMem-Bench, a multi-turn Text-to-SQL benchmark of 300 sessions and 1,400 turns built programmatically from three enterprise domains (BIRD financial, SEC EDGAR, Northwind), with deterministic ground truth and per-turn memory-critical annotation. We evaluate five frontier models -- GPT-5 mini, GPT-5.2, Claude Sonnet 4.5, Sonnet 4.6, and Opus 4.6 -- across five memory conditions enabling a three-way ablation isolating working-memory window size, episodic retrieval, and semantic augmentation as independent effects. All Claude models are evaluated with extended thinking enabled to maintain parity with GPT reasoning models. We introduce the Memory Benefit Score (MBS) as a per-turn diagnostic metric. Four findings emerge: (1) stateless multi-turn Text-to-SQL collapses to zero execution accuracy by Turn 3 across all five models, even under reasoning; (2) memory-architecture complexity does not monotonically improve accuracy -- working memory dominates, and additional components produce model- and dataset-dependent effects from +14 to -16 percentage points; (3) Claude Sonnet 4.6 underperforms Sonnet 4.5 by 17-33pp on SEC EDGAR across conditions, a generational regression persisting under reasoning; (4) under reasoning, Claude error distributions become mono-modal -- every non-correct turn is a wrong-result error. We release the benchmark, agent, and evaluation code.

Large Language Model interfaces are increasingly verbose, exposing intermediate reasoning traces alongside final answers. Traces are framed as transparency mechanisms, yet it is unclear how people use them to solve problems. We report a preregistered between-subjects study (N = 559) in which participants solved ten LSAT-style reasoning problems under one of three conditions: an Answer-only baseline, a Full-trace revealed before the answer, and a Summary-trace presented alongside the answer. Summaries preserved task performance at the no-trace baseline while significantly elevating trust and hedonic appeal, establishing that trace exposure shifts subjective appraisal of the interaction without bringing performance benefits. Under an open-weight reasoning model exposing verbose intermediate output, full traces additionally impaired performance relative to the answer-only baseline. Across all conditions, participants substantially overestimated their performance, and no trace format supported calibrated self-evaluation. Further analysis indicates that hedonic appeal, not trust, carries the indirect path to overestimation, consistent with a processing-fluency account. Reasoning traces are best understood as user-facing interface artifacts rather than transparent windows into model cognition, and calibration is unlikely to emerge from the traces themselves and may best be scaffolded by interactions that elicit users' own reasoning first.

Large Vision-Language Models (LVLMs) extend large language models with visual understanding, but remain vulnerable to hallucination, where outputs are fluent yet inconsistent with images. Recent studies link this issue to language bias-the tendency of LVLMs to over-rely on text while neglecting visual inputs. Yet most analyses remain empirical without uncovering its underlying cause. In this paper, we provide a systematic study of language bias and identify its root in modality misalignment during training. Our analysis shows that both Visual Instruction Tuning (VIT) and Direct Preference Optimization (DPO) often prioritize textual improvements, which may cause LVLMs to overly lean toward language modeling rather than balanced multimodal understanding. To address this, we propose two simple yet effective methods: Language Bias Regularization (LBR) which mitigates language bias through regularization during instruction tuning, and Language Bias Penalty (LBP), which penalizes language bias in the DPO training process. Extensive experiments across diverse models and benchmarks demonstrate the effectiveness of our approach. LBR consistently improves performance on over ten general benchmarks, while LBP significantly reduces hallucination and improves trustworthiness. Together, these methods not only mitigate language bias but also advance the overall alignment of LVLMs, all without introducing any additional data or auxiliary models. Our code is publicly available at https://github.com/lab-klc/LVLM-Language-Bias.

The growing use of Large Language Models (LLMs) for automated code generation has enhanced software development efficiency, but often at the cost of security. Generated code frequently overlooks critical concerns, leaving it vulnerable to issues such as weak encryption and improper input validation. To investigate this problem, we present a comprehensive empirical evaluation of the security quality of LLM-generated code across five LLMs and four programming languages (Java, C++, C, and Python), examining the impact of multiple prompt engineering methods. We introduce a weaknesses-aware zero-shot chain-of-thought (WA-0CoT) prompting strategy that enriches prompts with security context using CWE mappings to guide model reasoning. Our empirical analysis, supported by chi-square tests, finds no statistically significant reductions in vulnerability frequency or density across prompt methods. However, prompting strategies, including WA-0CoT, systematically influence the compositional distribution of CWE categories, with effects varying by programming language. These findings suggest that while security-aware prompting alters the structure of generated weaknesses, prompt engineering alone is insufficient to reliably reduce overall vulnerability levels. The results highlight the importance of language-aware and model-aware prompt design when evaluating the security properties of LLM-generated code.

Collaboration is the defining mode of modern science, yet its core mechanism -- feedback -- remains hard to observe, difficult to scale, and unequally distributed. Here we test whether large language models (LLMs) can contribute to this hidden but vital practice and reallocate scientific feedback, an essential yet scarce resource for knowledge production. In a global large-scale randomized field experiment, we delivered customized LLM-generated feedback for over 31,000 arXiv preprints across 150 fields and more than 45,000 researchers from 133 geographic regions. Relative to controls, authors who received feedback had a significantly higher likelihood of revising their manuscripts, corresponding to a 12.55% relative increase over the baseline revision rate. Exposure to AI feedback also increased authors' subsequent use of LLM tools in their future papers, suggesting longer-run shifts in scientific practice. These effects were strongest among authors from non-English-dominant research regions, manuscripts less embedded in the scholarly literature, and teams with lower h-indexes and earlier career stages, consistent with the idea that AI feedback may provide the greatest benefit where access to timely critique is otherwise limited. Together, these findings provide causal evidence that structured AI-based interventions can transform access to scientific feedback from a largely private advantage into a more widely distributed resource, with broader implications for productivity, equity, and capacity across the global research system.

Large Language Models (LLMs) are increasingly used to generate summaries of software bug reports, including sections such as Steps-to-Reproduce (S2R), Actual Behavior (AB), and Expected Behavior (EB). However, these models frequently produce hallucinations that can be convincing but unsupported by the source report. This can mislead developers and reduce trust in automated maintenance tools. Existing hallucination detection approaches typically evaluate outputs at the full-response level and do not consider the structure of technical documents. An initial exploratory study on 80 structured bug report summaries found that approximately 47.9% contained missing information, while 12.3% included fabricated content, highlighting the need for systematic hallucination analysis in bug report summarization. In this work, we empirically investigate hallucinations in LLM-generated bug report summaries from a section-aware perspective. Using the BugsRepo dataset, derived from Mozilla OSS projects, we introduce controlled synthetic hallucination injection to construct a benchmark for training and evaluation. We propose a section-aware hallucination detection approach that jointly predicts whether a summary contains hallucinated content, identifies affected sections, and classifies hallucination types. Experimental results across multiple pretrained language models show that the proposed approach achieves strong performance across all tasks, with the best model obtaining 0.89 report-level Macro-F1, 0.83 section-level Macro-F1, and 0.84 hallucination-type Macro-F1. We further analyze common hallucination patterns and model failure modes to better understand limitations of current LLM-generated bug report summaries. The findings highlight the importance of section-aware hallucination analysis for improving the reliability of LLM-assisted bug report summarization in software maintenance workflows.

Language agents increasingly improve by reusing \emph{skills} -- structured procedural artifacts distilled from past experience. In particular, \emph{domain-level} and \emph{model-generated} skills are especially promising. They offer fast adaptation within a domain by encoding domain-specific recurring procedures, and they scale beyond labor-intensive hand-crafting. However, while extraction methods continue to proliferate, understanding remains limited, with no comprehensive study spanning the full skill lifecycle -- \textbf{experience generation}, \textbf{skill extraction}, and \textbf{skill consumption} -- to ask whether such skills actually work, when they work, and what makes them succeed or fail. To close this gap, we build a utility-grounded evaluation framework that provides systematic experimental results across extractors and target agents, covering five diverse agentic task domains. We find that model-generated skills are beneficial on average but exhibit non-trivial negative transfer, and that neither extractors nor targets behave uniformly. A model can be a strong extractor yet a weak consumer, or vice versa, with skill utility independent of model scale or baseline task strength. To explain these patterns, we then dissect each lifecycle stage in depth, analyzing how experience composition shapes skill quality, what properties characterize useful skills, and how the same skill transfers across different consumers. Finally, we translate these findings into a concrete \emph{meta-skill} that guides skill extraction toward the features tied to actual utility, which consistently improves skill quality across domains and substantially reduces negative transfer.

AI-assisted code review tools typically operate as generic "expert reviewer" agents, producing homogeneous findings regardless of the analysis type needed. We present a system that constrains AI reviewer behavior through philosophical dispositions -- coherent personality lenses grounded in specific epistemological traditions (Pyrrhonist Skepticism, Navya-Ny=aya logic, Diogenes' Cynicism, Confucian relational ethics) that direct attention to structurally different types of issues. Each disposition is defined apophatically (by what it refuses to do), equipped with a self-monitoring failure mode (hamartia), and orchestrated in sequence by role protocols. We evaluate this system on 50 merged pull requests across 7 repositories spanning 5 programming languages (Python, Go, C++, Java, Terraform), 5 organizations (2 enterprise, 3 open-source), and 2 temporal eras (pre-AI 2020, post-AI 2024--2026). The disposition system achieves 46% convergence with human reviewers (validating signal quality), identifies unique findings at a 75% rate, and produces no findings judged false-positive by the author across 601 total findings (inter-rater agreement was not assessed and remains a limitation). A controlled baseline comparison demonstrates that 51% of disposition findings are not produced by the same model using generic "expert reviewer" prompting, and these unique findings target structural, operational, and logical concerns rather than standard code-level issues. Preliminary cross-model validation (Claude Opus vs.\ GPT Codex 5.3-xhigh) on 3 PRs shows 100% framework-structure adherence with 39% finding-level agreement, suggesting the framework provides real behavioral constraint while preserving model-specific analytical perspective.

Large Language Models (LLMs) are increasingly integrated into software systems for diverse purposes, due to their versatility, flexibility, and ability to simulate human reasoning to some extent. However, poor integration of LLM inference in source code can undermine software system quality. Therefore, inadequate LLM integration coding practices must be documented to help developers mitigate such issues. Following our earlier work on LLM code smells, this paper consolidates and refines the concept by presenting a self-contained taxonomy and a catalog of nine LLM code smells. We also create SpecDetect4LLM, a static source code analysis tool for their detection, and conduct extensive empirical evaluations of its detection effectiveness (precision and recall) as well as the prevalence of LLM code smells across 692 open-source software projects (171,194 source files). Our results show that LLM code smells affect 73.5% of the analyzed systems, with a detection precision of 91.3% and a recall of 71.8%.

Detecting Schwartz values in political texts is hard: cues are often implicit, and neighboring values differ by fine distinctions. Two remedies are widely assumed to help: more surrounding document text, and explicit moral knowledge. Knowledge-based retrieval has improved benchmarks elsewhere, but whether either transfers here is untested, because published systems vary context, knowledge, and model family at once. We separate these factors under matched conditions on the ValuesML/Touché ValueEval format. The input ranges from the target sentence to a local window to the full document. Retrieval is either absent or drawn from a curated moral knowledge base, injected by early, late, or cross-attention fusion. Supervised DeBERTa-v3 encoders are compared against zero-shot LLMs from 12B to 123B. More context is not uniformly better: full-document input improves the encoders by 2.5-3.8 macro-F1 points but does not consistently help the LLMs. Retrieved knowledge helps more reliably, improving every model family and context under early fusion. A control substituting random knowledge-base entries shows the families gain differently: LLMs from relevance, encoders from exposure to the value ontology. Neither larger encoders nor larger LLMs guarantee gains, and early fusion outperforms both trainable variants. Value-sensitive NLP should evaluate context, knowledge, and model family jointly.

Linear properties are ubiquitous in the representations of language models; however, testing them experimentally remains a challenging task. This work focuses on relational linearity: the hypothesis that, for a fixed relation (e.g., "plays"), the unembedding of an object (e.g., "trumpet") can be predicted from the embedding of its subject (e.g.,"Miles Davis") by a linear map. We present an experimental method to test the formulation of relational linearity by Marconato et al. (2025). Specifically, we introduce a probing method, based on Kullback-Leibler divergence, to evaluate this property and examine its variation across layers and paraphrased relational queries. It is also more efficient than previous work; for example, it avoids the crude Jacobian approximations used in Linear Relational Embeddings by Hernandez et al. (2024). Our findings across four datasets show that relational linearity varies across models, exhibits layer-wise patterns consistent with prior observations about linguistic information in model representations, and is differently affected by changes in how the relation is phrased.

Third-party Python libraries introduce dependency management overhead, supply chain risk, and deployment friction in constrained environments. A natural question is how much of this ecosystem can be replicated using only Python's standard library -- and at what correctness and performance cost. We address this empirically through zerodep, a growing collection of single-file Python modules, each a stdlib-only reimplementation of a popular third-party library, developed with LLM assistance under strict constraints: no external imports, single file, drop-in API compatibility, and mandatory correctness validation against the reference library. Spanning over 40 modules across 12 categories -- including serialization, networking, cryptography, agent protocols, and text processing -- zerodep provides a controlled testbed for two interrelated questions: (1) Where does the stdlib suffice? and (2) Can LLMs effectively generate correct, performant code under tight symbolic constraints? Systematic benchmarking shows that stdlib-only implementations achieve performance parity (within 2x of the reference) in the majority of cases. The primary performance cliff is C-extension-backed computation (image processing, binary serialization, low-level crypto), not the inherent overhead of pure-Python third-party libraries. Conversely, many widely-used libraries carry architectural overhead that LLM-generated stdlib reimplementations avoid, yielding 5--115x speedups in several categories. We characterize the stdlib capability boundary across complexity tiers and library categories, discuss where LLM-assisted development succeeds and where it requires iterative human correction, and examine implications for dependency-free software engineering at scale. zerodep is open-source at https://github.com/Oaklight/zerodep.

Feedforward steering control is a key component of hierarchical control architectures for autonomous racing. The goal is to reduce steering corrections from the feedback controllers by predicting the vehicle's inverse lateral dynamics. This paper presents a systematic benchmark of two learning-based and two empirical (analytical) feedforward steering controllers. We introduce a new \acf{ehd} formulation based on a polynomial surface fit that captures velocity-dependent nonlinear steering behavior with minimal parametrization. We test the feedforward controllers in a high-fidelity simulation framework based on the real-world Abu Dhabi Autonomous Racing League competition, using a high-fidelity double-track vehicle dynamics simulator. Open-loop evaluation shows that the learning-based controllers achieve the lowest prediction errors; however, closed-loop testing reveals that this improved accuracy does not translate into superior path tracking performance or lap times, even after iterative fine-tuning. In contrast, the proposed EHD approach achieves the best overall closed-loop robustness and lap time, highlighting the necessity of evaluating feedforward strategies within the complete trajectory planning and control software stack. Our code is available at https://github.com/TUMRT/steering_ff_control.

Quadratic constraints (QCs) are widely used to characterize nonlinearities and uncertainties, but generic analytical characterizations can be conservative on bounded domains. This paper develops a framework for constructing verified quadratic characterizations of scalar relations in the two-dimensional real plane. Candidate quadratic inequalities are locally generated by solving convex quadratic programs using samples from the relation and exterior sample points. They are then verified globally using sum-of-squares certificates over an exact semialgebraic description or, in the case of nonpolynomial relations, over relaxed polynomial descriptions. The resulting verified constraints define a sound overapproximation of the scalar relations over the considered domains. These constraints are directly compatible with existing analysis frameworks based on QCs and pointwise integral quadratic constraints (IQCs) for static nonlinearities and uncertainties, and they can also be embedded in QC-based semidefinite programs for reachability and safety analysis of feedforward neural networks. For smooth activations such as $\tanh$, the method yields domain-dependent quadratic characterizations that constitute an alternative to generic sector- or slope-based descriptions. For ReLU networks, we give methods to reduce conservatism in QC-based reachability analysis of feedforward networks by exploiting dependencies between neurons and tighter local bounds. Numerical examples demonstrate improved reachability results for smooth activations, reduced conservatism for ReLU networks, and applicability beyond neural networks through an example involving saturation.

Modern deployments of Large Language Models (LLMs) increasingly require serving multiple models with diverse architectures, sizes, and specialization on shared, heterogeneous hardware. This setting introduces new challenges for resource allocation, dispatching, and scheduling, particularly under GPU memory constraints where partial CPU-GPU offloading and preemption become necessary. While existing systems primarily optimize throughput for a single model, comparatively little work addresses multi-model scheduling under these conditions. In this paper, we present an empirical study of how different LLMs behave across hardware platforms, focusing on the performance implications of layer offloading and preemption. We show that offloading leads to strongly non-linear and model-dependent degradation in decode throughput, with smaller models exhibiting sharper sensitivity to reduced GPU residency. We further demonstrate that preemption incurs substantial overhead, largely dominated by model state reload rather than key-value cache transfer, and that this cost varies significantly across models and hardware platforms. Additionally, we highlight the role of sequence length and interconnect bandwidth in amplifying data movement and execution inefficiencies. Based on these findings, we identify a set of key features that future schedulers must consider, including model-specific offloading sensitivity, workload characteristics, and the cost structure of preemption and data transfer. These insights provide guidance for the design of next-generation LLM serving systems capable of efficiently managing heterogeneous, multi-model workloads with hybrid CPU-GPU execution.

Kolmogorov-Arnold Networks (KANs) have demonstrated an exceptional ability to learn complex functions on clean, low-dimensional data but struggle to maintain performance on noisy and imperfect real-world datasets. In contrast, conventional multi-layer perceptrons (MLPs) are far more tolerant to noise and computationally efficient. Replacing all MLP components with KANs in HAR models often degrades accuracy and computation efficiency, highlighting an open challenge: how to combine KANs' precision with MLPs' noise robustness and efficiency. To address this, we systematically explore various placements of KAN modules within deep HAR networks and propose a hybrid architecture that strategically synergizes the strengths of both paradigms, which uses a KAN-based input embedding layer, retains MLP layers for intermediate feature mixing, and introduces a specialized LarctanKAN module for final activity classification. Across eight public HAR datasets, the hybrid KAN-MLP model achieves an average macro F1 score relative improvement of 5.33% compared pure-MLP model, significantly outperforming standalone KAN and MLP baselines. Furthermore, integrating this hybrid strategy into other state-of-the-art HAR architectures consistently boosts their performance. Our findings demonstrate that a carefully orchestrated combination of KAN, MLP, or other conventional neural components yields more robust and accurate HAR models for real-world wearable sensing environments.

Prior work on fine-grained image recognition (FGIR) has established the importance of the backbone selection, but has neglected the accuracy-vs-cost trade-offs under different training and evaluation settings. In this work we conduct a large-scale study with over 2000 experiments across 6 training and evaluation settings, 9 pretrained backbones, and 17 datasets. Preliminary observations on the effectiveness of data augmentation for fine-grained training motivate us to extend Counterfactual Attention Learning (CAL), a state-of-the-art method based on data-aware cropping and masking augmentations, with cross-image discriminative region mixing augmentation. We also propose an efficient evaluation-only variant that maintains competitive accuracy while reducing inference costs by forfeiting the forward pass on discriminative crops that is normally used by CAL and similar FGIR methods. Our results show that data-aware augmentations during training only can enable a model to achieve excellent accuracy even without crops, significantly reducing inference costs. To support future research we share our code and checkpoints at: \url{https://github.com/arkel23/FGIR-Backbones}

LLM-based chatbot agents increasingly process user requests by combining natural-language reasoning with external tools such as web browsing. These capabilities improve usability, but they also create attack surfaces when untrusted external content is processed as part of a user' s task. This paper studies a privacy-leakage attack chain based on indirect prompt injection in black-box chatbot environments, where the attacker has no access to model weights, system prompts, or agent implementation details including how a trajectory is actually managed during its processing for a query. We first analyze how an attacker can hijack an agent' s intended task by crafting external content that appears benign to the victim while inducing the agent to execute an attacker-defined objective. We then evaluate a new prompt-injection technique, called exemplification, which uses a bridge in the external content to reframe the user prompt and the benign beginning of the retrieved page as few-shot examples before appending the attacker' s objective. We compare its attack success rate with a prior fake-completion technique. Finally, we demonstrate a proof-of-concept data-exfiltration chain using fictitious personal information in a controlled setting. Our results suggest that prompt injection, jailbreak-style instruction steering, and web-tool invocation can be combined into a feasible privacy-leakage path in deployed chatbot agents.

Joint-Embedding Predictive Architectures (JEPA) are a promising framework for self-supervised video representation learning, yet the behavior of auxiliary objectives in small-scale Video-JEPA training is not well characterized. We report a small-scale empirical study of 18 auxiliary objective variants for Video-JEPA across two pretraining regimes: single-dataset (UCF-101) and mixed-dataset (UCF-101 + Something-Something V2 + ImageNet-100). We evaluate frozen representations on three complementary benchmarks: Diving-48 (fine-grained motion), SomethingSomething V2 (temporal reasoning), and ImageNet-100 (appearance). Our experiments suggest that many auxiliary objectives exhibit capacity trade-offs: gains on one downstream capability often coincide with degradation on another. We then study FWM-HW-LD (Factorized World-Model with Hard-Region-Weighted Latent Dynamics), a training-time objective that separates the latent representation into appearance and dynamics subspaces and applies hard-region weighting to both JEPA prediction errors and latent dynamics errors. In our mixed-dataset setting, FWM-HW-LD improves ImageNet-100 by +5.92 and SSv2 by +3.21 percentage points relative to the reference baseline, while remaining within 0.30 percentage points on Diving-48. These results indicate that latent factorization is a useful direction for studying auxiliary-objective trade-offs in Video-JEPA.

Recent advances in Time Series Foundation Models (TSFMs) promise zero-shot forecasting capabilities with minimal task-specific training. While these models have shown strong performance across generic benchmarks, their applicability in volatile, complex electricity markets remains underexplored. Addressing this gap, this study provides a systematic empirical evaluation of several TSFMs, specifically Chronos-2 and Chronos-Bolt (developed by Amazon), and TimesFM 2.5 (provided by Google), for forecasting Belgian day-ahead and imbalance electricity prices. For both considered markets, Chronos-2 in ARX mode produces the most accurate forecasts. Compared with the best ensemble prediction from other machine learning methods, Chronos-2's Mean Absolute Error (MAE) is 5% lower for the day-ahead market. In contrast, the model yields 10% higher MAE predicting imbalance prices across all forecast horizons, except for the two-hour-ahead horizon. Moreover, we find that TSFMs exhibit genuine zero-shot forecasting skills but still struggle under extreme market conditions.

The rapid evolution of Large Language Model (LLM) agents has produced diverse interaction paradigms, yet few production systems integrate multiple paradigms within a unified architecture. This paper presents a systematic analysis of three principal agent interaction paradigms, including Multi-Agent Orchestration (Generator-Evaluator), ReAct Tool-Use Loops, and Memory-Augmented Interaction, as implemented in buddyMe, an open-source multi-model agent programming framework. We formalize a five-stage processing pipeline: Requirement Pre-Review -> Task Decomposition -> ReAct Execution -> Real-Execution Verification -> Adversarial Evaluation Discussion, and establish a six-dimensional evaluation schema with weighted scoring. Through four empirical case studies drawn from real-world deployment logs covering museum guide generation, scheduled weather tasks, and comprehensive tour planning, we draw three key conclusions. First, Generator-Evaluator pre-review detects requirement omissions in 20 percent of complex tasks, with 80 percent tasks passing initial inspection. Second, the ReAct loop ensures stable subtask execution but leads to around 30 percent redundant tool invocations. Third, adversarial Evaluator-Defender discussions reach consensus within 2-3 rounds for nearly 70 percent of scenarios, functioning mainly for content refinement rather than logical reversal. We additionally provide three Mermaid-based architectural diagrams and conduct cross-paradigm comparisons with CrewAI, AutoGen, LangGraph, MemGPT and A-Mem across six system dimensions. The research outcomes offer practical design guidelines for constructing stable and reliable multi-paradigm agent systems.

Despite the remarkable success of transformer architectures in natural language processing, their scalability limitations remain poorly understood through systematic empirical analysis. This paper presents the first comprehensive large-scale evaluation of 118 transformer models across seven distinct architectural categories, revealing fundamental performance walls that manifest as hard deployment constraints. Our systematic benchmarking methodology uncovers a critical scalability crisis: while 88.1% of models successfully process sequences up to 512 tokens, this drops dramatically to 44.9% at 1024 tokens, with complete failure (0%) at 2048 tokens. Through rigorous analysis of loading times, memory consumption, and computational efficiency across sequence lengths from 128 to 2048 tokens, we demonstrate that compressed models achieve superior parameter efficiency (649.2 tokens/sec/M parameters) compared to large generative models (12.5 tokens/sec/M). Our findings challenge prevailing scaling assumptions and provide the first quantitative evidence that the theoretical O(n2) attention complexity translates into measurable performance walls. This work establishes new benchmarking methodologies for transformer evaluation and provides critical insights for practical deployment decisions in production environments.

In recent years, computer vision has witnessed remarkable progress, fueled by the development of innovative architectures such as Convolutional Neural Networks (CNNs), Generative Adversarial Networks (GANs), diffusion-based architectures, Vision Transformers (ViTs), and, more recently, Vision-Language Models (VLMs). This progress has undeniably contributed to creating increasingly realistic and diverse visual content. However, such advancements in image generation also raise concerns about potential misuse in areas such as misinformation, identity theft, and threats to privacy and security. In parallel, Mamba-based architectures have emerged as versatile tools for a range of image analysis tasks, including classification, segmentation, medical imaging, object detection, and image restoration, in this rapidly evolving field. However, their potential for identifying AI-generated images remains relatively unexplored compared to established techniques. This study provides a systematic evaluation and comparative analysis of Vision Mamba models for AI-generated image detection. We benchmark multiple Vision Mamba variants against representative CNNs, ViTs, and VLM-based detectors across diverse datasets and synthetic image sources, focusing on key metrics such as accuracy, efficiency, and generalizability across diverse image types and generative models. Through this comprehensive analysis, we aim to elucidate Vision Mamba's strengths and limitations relative to established methodologies in terms of applicability, accuracy, and efficiency in detecting AI-generated images. Overall, our findings highlight both the promise and current limitations of Vision Mamba as a component in systems designed to distinguish authentic from AI-generated visual content. This research is crucial for enhancing detection in an age where distinguishing between real and AI-generated content is a major challenge.

Malformed, missing, or boundary-value inputs in microservice APIs can cascade across dependent services, threatening reliability. Robustness testing systematically exercises such inputs to expose server-side failures, but generating diverse, effective tests remains challenging. Large Language Models can generate such tests from API specifications; however, it is unknown whether different models and prompt strategies produce diverse failure sets or converge on the same failures. We report a controlled experiment applying 7 prompt strategies to 3 open-source LLMs (14B-70B parameters) targeting 2 architecturally distinct microservice systems: one Java monolingual (6 services, 9 failure modes) and one polyglot (27 services, 14 failure modes), yielding 38 valid runs and 663 generated tests. We find that prompt strategy explains more variation in diversity than model size: a Structured prompt collapses diversity entirely, while a single model varied across three prompt strategies achieves complete failure-mode coverage on one system, outperforming any multi-model ensemble under a fixed prompt. We introduce two strategies, Guided and GuidedFewShot, that embed a mutation taxonomy from prior robustness testing research as domain context. GuidedFewShot achieves the highest single-run coverage on both systems (5 of 9 and 8 of 14 failure modes) while maintaining low cross-model similarity. A key lesson is that taxonomy rules alone are insufficient: LLMs cannot distinguish key-absent from value-empty mutations without concrete examples. Findings replicate across both systems.

Iterative self-refinement is a simple inference-time strategy for machine translation: an LLM revises its own translation over multiple inference-time passes. Yet document-scale refinement remains poorly understood: 1) which pipelines work best, 2) what quality dimensions improve, and 3) how refiners behave. In this paper, we present a systematic study of document-level literary translation, covering nine LLMs and seven language pairs. Across nine translation-refinement granularity combinations and five refinement strategies, we find a robust recipe: document-level MT followed by segment-level refinement yields strong and stable improvements. In contrast, document-level refinement often makes fewer edits and leads to smaller or less reliable gains. Beyond granularity, A simple general refinement prompt consistently outperforms error-specific prompting and evaluate-then-refine schemes. Our large-scale human evaluation shows that refinement gains come primarily from fluency, style, and terminology, with limited and less consistent improvements in adequacy. Experiments varying model strength reveal refinement projects outputs toward the refiner's distribution rather than performing targeted error repair. These findings clarify the mechanisms and limitations of current refinement approaches.

Validating training data for reasoning models typically requires expensive trial-and-error fine-tuning cycles. In this work, we investigate whether the utility of a reasoning dataset can be reliably predicted prior to training using intrinsic data metrics. We propose a suite of quantitative measures and evaluate their predictive power by fine-tuning 8B and 11B models on semantically distinct variants of a Polish reasoning dataset. Our analysis reveals that these intrinsic metrics demonstrate strong and significant correlations with downstream model performance. Crucially, we find that the predictors of utility are scale-dependent: smaller models rely on alignment-focused metrics to ensure precision, whereas larger models benefit from high redundancy, utilizing verbose traces to solve complex tasks. These findings establish a scale-aware framework for validating reasoning data, enabling practitioners to select effective training sets without the need for exhaustive empirical testing.

Automated detection of vulnerability-fixing commits (VFCs) is critical for timely security patch deployment, as advisory databases lag patch releases by a median of 25 days and many fixes never receive advisories. We present a comprehensive evaluation of code language model based VFC detection through a unified framework consolidating over 20 fragmented datasets spanning more than 180000 commits. Across over 180 experiments with fine-tuned models from 125 M to 14 B parameters, we find no evidence that models acquire transferable security-relevant code understanding from code changes alone. When commit messages are available, they dominate model attention, and when removed, an attribution analysis shows that enriching diffs with additional intra-procedural semantic context does not shift model attention toward the code changes. Group-stratified evaluation exposes approximately 17% performance drops compared to random splits, while temporal splits on aggregated datasets prove unreliable due to compositional shift in the underlying project distributions. At a false positive rate of 0.5% all fine-tuned code-only models miss over 93% of vulnerabilities. Larger and more diverse training data or generative approaches show preliminary improvements but do not resolve the underlying limitations. To support future research on code-centric VFC detection, we release our unified framework and evaluation suite.

Large language models (LLMs) are increasingly used as substitutes for human subjects in behavioral simulations, including synthetic social network generation. Yet it remains unclear how their relational outputs depend on prompt design, cultural framing, prompt language, and model scale. Building on homophily theory and structural balance theory, we formalize four LLM-based tie-formation mechanisms: sequential, global, local, and iterative, and treat them as distinct conditional distributions over edge sets. Using a fixed roster of 50 demographically grounded personas, we generate 192 verified directed networks across four cultural contexts, four prompt languages, three GPT-4.1 variants, and four prompting architectures, with two seeds per condition. We find that cultural framing shifts inbreeding homophily and largest-component connectivity. Political affiliation dominates tie formation under three methods, while the global method substitutes age, showing that prompt architecture functions as a substantive sociological variable. Model scale produces a stable divergence ranking, with the smallest variant behaving qualitatively differently rather than merely noisily. Prompt language alone sharply shifts religion homophily, especially under Hindi prompting, while leaving political homophily nearly invariant. LLM-generated networks match real social graphs on clustering and modularity better than standard graph baselines, yet encode demographic biases above empirical levels. These results show that prompt choices often treated as implementation details encode substantive sociological assumptions.

Specialized foundation models are beginning to emerge in various medical subdomains, but pretraining methodologies and parametric scaling with the size of the pretraining dataset are rarely assessed systematically and in a like-for-like manner. This work focuses on foundation models for electrocardiography (ECG) data, one of the most widely captured physiological time series world-wide. We present a comprehensive assessment of pretraining methodologies, covering five different contrastive and non-contrastive self-supervised learning objectives for ECG foundation models, and investigate their scaling behavior with pretraining dataset sizes up to 11M input samples, exclusively from publicly available sources. Pretraining strategy has a meaningful and consistent impact on downstream performance, with contrastive predictive coding (slightly ahead of JEPA) yielding the most transferable representations across diverse clinical tasks. Scaling pretraining data continues to yield meaningful improvements up to 11M samples for most objectives. We also compare model architectures across all pretraining methodologies and find evidence for a clear superiority of structured state space models compared to transformers and CNN models. We hypothesize that the strong inductive biases of structured state space models, rather than pretraining scale alone, are the primary driver of effective ECG representation learning, with important implications for future foundation model development in this and potentially other physiological signal domains.

The literature on how large language models handle conflict between their training knowledge and a contradicting document presents a persistent empirical contradiction: some studies find models stubbornly retain their trained answers, ignoring provided documents nearly half the time, while others find models readily defer to the document, following context approximately 96% of the time. We argue these contradictions dissolve once one recognises that prior experiments have studied three qualitatively distinct processing situations without distinguishing them. We propose a three-regime framework: Regime 1 (single-source updating, dominant predictor: evidence coherence), Regime 2 (competitive integration, dominant predictor: parametric certainty), and Regime 3 (task-appropriate selection, dominant predictor: task knowledge requirement). We formalise a distinction between parametric strength (exposure frequency) and parametric uniqueness (encoding consistency), showing empirically that these are orthogonal dimensions (r = -0.002, p = .97) with strength as the operative predictor in stable factual domains. We validate the framework across Claude Sonnet 4.6, GPT-5.5, Gemini 2.5 Flash, Llama 4 Maverick, and DeepSeek V3 using 9,970 API calls in three experimental phases. GEE logistic regression confirms the predicted Regime 2 certainty gradient for all five models (beta = -0.38 to -0.50, all p <= .013, BH-FDR corrected). A Regime 3 ablation shows task framing alone flips context-following from near-100% (contextual knowledge condition) to 6-71% (parametric knowledge condition), with all five models significant (p < .001). The certainty gradient is robust to multinomial outcome modeling, sensitivity analyses for hedging responses, and FDR correction.

Agent evaluation requires assessing complex multi-step behaviors involving tool use and intermediate reasoning, making it costly and expertise-intensive. A natural question arises: can frontier coding assistants reliably automate this evaluation process? Our study shows that simply prompting coding assistants is insufficient for this task. Without domain-specific evaluation knowledge, frontier coding assistants achieve only a 30% execution success rate and produce over-engineered evaluations averaging 12+ metrics per agent, indicating that strong coding ability does not automatically translate to reliable agent evaluation. We introduce EvalAgent, an AI assistant that automates the end-to-end agent evaluation pipeline. EvalAgent encodes evaluation domain expertise as evaluation skills (procedural instructions, reusable code and templates, and dynamically retrieved API documentation) that compose into a trace-based pipeline producing complete evaluation artifacts including metrics, executable code, and reports. To systematically assess generated evaluations, we introduce a meta-evaluation framework alongside AgentEvalBench, a benchmark comprising 20 agents, each paired with evaluation requirements and test scenarios. We further propose the Eval@1 metric to measure whether generated evaluation code both executes and yields meaningful results on the first run. Our experiments show that EvalAgent produces focused evaluations, improving Eval@1 from 17.5% to 65%, and achieving 79.5% human expert preference over baseline approaches. Further ablation studies show that evaluation skills are critical for handling complex evaluation: removing them causes Eval@1 to drop significantly from 65% to 30%.

This work proposes Attractor-Vascular Coupling Theory (AVCT), a mathematical framework showing that cardiac attractor geometry encodes blood pressure (BP) information sufficient for AAMI-standard estimation, and validates the theory through a calibrated cuffless BP model using photoplethysmography (PPG). AVCT is grounded in Cardiac Stability Theory and operationalized using Takens delay embedding and attractor morphology extraction. Two theorems, one proposition, and one corollary formally justify the use of PPG attractor features for BP estimation and predict the feature-importance hierarchy. A LightGBM model trained on pulse transit time (PTT) and Cardiac Stability Index (CSI) attractor features under single-point calibration was evaluated using strict leave-one-subject-out cross-validation (LOSO-CV) on 46 subjects from BIDMC ICU (n = 9) and VitalDB surgical data (n = 37), comprising 29,684 windows. The model achieved systolic BP (SBP) mean absolute error (MAE) of 2.05 mmHg and diastolic BP (DBP) MAE of 1.67 mmHg, with correlations r = 0.990 and r = 0.991, satisfying the AAMI/IEEE SP10 requirement of MAE below 5 mmHg. Median per-subject MAE was 1.87/1.54 mmHg, and 70%/76% of subjects individually satisfied AAMI criteria. A PPG-only ablation using nine smartphone attractor features matched the ECG+PPG model within 0.05 mmHg, demonstrating that clinical-grade BP tracking is achievable using only a smartphone camera while surpassing prior generalized LOSO-CV results using fewer sensors. All four AVCT predictions were quantitatively confirmed, with 91.5% error reduction from uncalibrated to calibrated estimation (epsilon_cal = 0.915). Unlike post-hoc explainable AI methods, AVCT predicts features satisfying the architectural faithfulness criterion of the Explainable-AI Trustworthiness (EAT) framework and grounding BP estimation in nonlinear dynamical systems theory.

Large Language Models(LLMs) are increasingly explored for cybersecurity applications such as vulnerability detection. In the domain of threat modelling, prior work has primarily evaluated a number of general-purpose Large Language Models under limited prompting settings. In this study, we extend the research area of structured threat modelling by systematically evaluating domain-adapted language models of different sizes to their general counterparts. We use both LLMs and Small Language Models(SLMs) that were domain adapted to telecommunications and cybersecuirty. For the structured threat modelling, we selected the widely used STRIDE approach and the application area is 5G security. We present a comprehensive empirical evaluation using 52 different configurations (on 8 different language models) to analyze the impact of 1) domain adaptation, 2) model scale, 3) decoding strategies (greedy vs. stochastic sampling), and 4) prompting technique on STRIDE threat classification. Our results show that domain-adapted models do not consistently outperform their general-purpose counterparts, and decoding strategies significantly affect model behavior and output validity. They also show that while larger models generally achieve higher performance, these gains are neither consistent nor sufficient for reliable threat modelling. These findings highlight fundamental limitations of current LLMs for structured threat modelling tasks and suggest that improvements require more than additional training data or model scaling, motivating the need for incorporating more task-specific reasoning and stronger grounding in security concepts. We present insights on invalid outputs encountered and present suggestions for prompting tailored specifically for STRIDE threat modelling.

Current Large Multimodal Models (LMMs) struggle with spatial reasoning tasks requiring viewpoint-dependent understanding, largely because they are confined to a single, static observation. We propose Thinking with Novel Views (TwNV), a paradigm that integrates generative novel-view synthesis into the reasoning loop: a Reasoner LMM identifies spatial ambiguity, instructs a Painter to synthesize an alternative viewpoint, and re-examines the scene with the additional evidence. Through systematic experiments we address three research questions. (1) Instruction format: numerical camera-pose specifications yield more reliable view control than free-form language. (2) Generation fidelity: synthesized view quality is tightly coupled with downstream spatial accuracy. (3) Inference-time visual scaling: iterative multi-turn view refinement further improves performance, echoing recent scaling trends in language reasoning. Across four spatial subtask categories and four LMM architectures (both closed- and open-source), TwNV consistently improves accuracy by +1.3 to +3.9 pp, with the largest gains on viewpoint-sensitive subtasks. These results establish novel-view generation as a practical lever for advancing spatial intelligence of LMMs.

The real world unfolds along a single set of physics laws, yet human intelligence demonstrates a remarkable capacity to generalize experiences from this singular physical existence into a multiverse of games, each governed by entirely different rules, aesthetics, physics, and objectives. This omni-reality adaptability is a hallmark of general intelligence. As Artificial Intelligence progresses towards Artificial General Intelligence, the multiverse of games has evolved from mere entertainment into the ultimate ground for training and evaluating AGI. The pursuit of this generality has unfolded across four eras: from environment-specific symbolic and reinforcement learning agents, to current large foundation models acting as generalist players, and toward a future creator stage where agent both creates new game worlds and continually evolves within them. We trace the full lifecycle of a generalist game player along four interdependent pillars: Dataset, Model, Harness, and Benchmark. Every advance across these pillars can be read as an attempt to break one of five fundamental trade-offs that currently bound the whole system. Building on this end-to-end view, we chart a five-level roadmap, progressing from single-game mastery to the ultimate creator stage in which the agent simultaneously creates and evolves within theoretical game multiverse. Taken together, our work offers a unified lens onto a rapidly shifting field,and a principled path toward the omnipotent generalist agent capable of seamlessly mastering any challenge within the multiverse of games, thereby paving the way for AGI.

Tool-enabled AI agents are increasingly deployed in cloud-hosted environments and offered as services, where they perform side-effecting operations through privileged tools within execution environments. While such agents enable powerful automation, the security implications of hosting autonomous agents in privileged execution environments are not yet fully explored. This paper presents a structured analysis of security risks associated with cloud-hosted AI agents. We introduce a taxonomy of risk categories, illustrate these risks through three representative agent scenarios, and discuss mitigation strategies along with their tradeoffs. A small controlled experiment empirically illustrates risk manifestation and the effect of lightweight mitigations in this setup. Our analysis suggests that many risks in autonomous cloud agents arise not from novel vulnerabilities, but from over-privileged tools, capability-intent mismatches, and ambient authority leakage in execution environments. Based on these findings, we derive practical design guidelines for deploying AI agents in the cloud more securely.

As generative AI and low-code workflow platforms become routine in software practice, a key educational question is whether the next generation of computer engineers will accept these tools as useful, usable, and worthy of sustained engagement. This paper reports a mixed-methods, cross-sectional study of undergraduate computer engineering students' acceptance of AI automation tooling, instantiated through the open-source platform n8n across three identically scripted workshops in Thailand (n = 103). A 12-item, five-point Likert instrument mapped to six TAM/UTAUT constructs - Performance Expectancy (PE), Effort Expectancy (EE), Behavioral Intention (BI), Self-Efficacy (SE), Hedonic Motivation (HM), and Output Quality (OQ) - was complemented by inductive thematic analysis of open-ended feedback. Analyses combined ordinal reliability estimation, bootstrap confidence intervals, non-parametric tests, multiple-comparison-controlled correlations, polychoric dimensionality diagnostics, a common-method-bias check, and between-session comparisons. Acceptance was favorable across all six constructs with large effect sizes, with PE emerging as the strongest construct and HM as the weakest. Dimensionality diagnostics further revealed that canonical TAM/UTAUT sub-facets collapsed into a single general acceptance factor in this short-form post-workshop context, a finding with important methodological and theoretical implications. Qualitative themes converged with the quantitative profile regarding usefulness and enthusiasm but diverged on output quality, revealing a small yet articulate reliability-skeptical minority. The findings support the curricular adoption of AI automation tooling in undergraduate computing education and identify three theory-grounded instructional levers: instruction-sequencing scaffolds, self-efficacy supports, and trust-calibration interventions.

Large language models (LLMs) are known to generate plausible but false information across a wide range of contexts, yet the real-world magnitude and consequences of this hallucination problem remain poorly understood. Here we leverage a uniquely verifiable object - scientific citations - to audit 111 million references across 2.5 million papers in arXiv, bioRxiv, SSRN, and PubMed Central. We find a sharp rise in non-existent references following widespread LLM adoption, with a conservative estimate of 146,932 hallucinated citations in 2025 alone. These errors are diffusely embedded across many papers but especially pronounced in fields with rapid AI uptake, in manuscripts with linguistic signatures of AI-assisted writing, and among small and early-career author teams. At the same time, hallucinated references disproportionately assign credit to already prominent and male scholars, suggesting that LLM-generated errors may reinforce existing inequities in scientific recognition. Preprint moderation and journal publication processes capture only a fraction of these errors, suggesting that the spread of hallucinated content has outpaced existing safeguards. Together, these findings demonstrate that LLM hallucinations are infiltrating knowledge production at scale, threatening both the reliability and equity of future scientific discovery as human and AI systems draw on the existing literature.

Qualitative analysis plays a pivotal role in understanding the human and social aspects of software engineering. However, it remains a demanding process shaped by the subjective interpretation of individual researchers and sensitive to methodological choices such as prompt design. Recent advancements in Large Language Models (LLMs) offer promising opportunities to support this type of analysis, although their reliability in reproducing human qualitative reasoning under varying prompting conditions remains largely untested. This study presents a controlled empirical evaluation of three LLMs -- Claude Haiku, DeepSeek-Chat, and Gemini 2.5 Flash -- across two prompt engineering strategies (zero-shot and multi-shot closed coding), using Cohen's kappa as the primary agreement metric over ten independent runs per configuration. Results suggest that multi-shot prompting significantly improves agreement for Claude Haiku (Delta kappa = +0.034, Wilcoxon p = 0.004) but not for DeepSeek-Chat or Gemini 2.5 Flash. Intra-model stability varies substantially -- DeepSeek-Chat and Claude Haiku exhibit the lowest variance (SD approx. 0.017), while Gemini 2.5 Flash is the least stable (SD = 0.038). A systematic over-prediction of "Sharing Negative Feedback" is identified across all models (bias ratios up to 5.25x), alongside consistent under-prediction of "Expressing Concerns." Collectively, these findings provide empirical evidence for prompt engineering guidelines in LLM-assisted qualitative coding for software engineering research.

Adaptive rank allocation for LoRA, allocating more parameters to important layers and fewer to unimportant ones, consistently improves efficiency under supervised fine-tuning (SFT). We investigate whether this success transfers to reinforcement learning, specifically Group Relative Policy Optimization (GRPO). Using gradient-magnitude profiling on Qwen 2.5 1.5B with GSM8K, we find that it does not: proportional rank allocation degrades accuracy by 4.5 points compared to uniform allocation (70.0% vs. 74.5%), despite using identical parameter budgets. We identify two mechanisms behind this failure. First, the gradient landscape under GRPO is fundamentally flatter than under SFT, the max-to-min layer importance ratio is only 2.17x, compared to >10x reported in SFT literature. All layers carry meaningful gradient signal; none are truly idle. Second, we discover a gradient amplification effect: non-uniform allocation widens the importance spread from 2.17x to 3.00x, creating a positive feedback loop where high-rank layers absorb more gradient while low-rank layers are progressively silenced. Our results suggest that gradient importance does not predict capacity requirements under RL, and that naive transfer of SFT-era rank allocation to alignment training should be avoided.

The evolution of generative models from next-token predictors to autonomous engines of complex systems necessitates rigorous safety hardening. Adversarial jailbreaking, the strategic manipulation of models to elicit harmful output, remains a primary threat to safe deployment. While Reinforcement Learning (RL) frames jailbreaking as a multi-step attack through sequential optimization, a mechanistic understanding of why the framework succeeds remains incomplete. To fill this gap, we present the first systematic decomposition of RL jailbreaking. We deconstruct the framework into problem formalization (reward function, action space, episode length), and algorithmic measures (RL algorithm, training data, reward-shaping) to identify the structural determinants of adversarial success. Our results reveal that the RL-jailbreaker successfully compromised all targeted models and safeguards. Through this first-of-its-kind analysis, we demonstrate that environment formalization, specifically dense rewards and extended episode lengths, is the primary driver of jailbreaking success. This work provides a tool for improving RL-jailbreaker efficiency and, ultimately, harden generative models resistant to RL-based attacks.

Large Language Models (LLMs) have achieved great improvements in recent years. Nevertheless, it still remains unclear how good LLMs are for reasoning tasks, especially for long-chain ones. In this paper, we evaluate LLMs' performance on the simplest yet long-chain reasoning task, namely the Equivalence Class Problem (ECP), i.e., determining whether two variables are equal given a set of randomly generated equivalence relations. We consider both reasoning and non-reasoning representative LLMs over a large variety of problem instances, ranging over different numbers of variables, connectivity probabilities, prompts, and other factors. The experimental results show that non-reasoning LLMs fail ECP, while reasoning models are significantly better but still struggle to completely solve this problem. Interestingly, considering various connectivity probabilities with a fixed number of variables, we observe that, for non-reasoning models, the hardest problem instances coincide with the phase transition point of ln n/(n-1), suggesting the chaos of the problem; in contrast, for reasoning models, the hardest ones coincide with the biggest diameter, suggesting the reasoning difficulty of the problem.

Large language models (LLMs) exhibit a wide range of human-like behaviors, from expressing thoughts and emotions, to engaging in relationship-building with users, to refusing requests and maintaining boundaries. Despite their prevalence, researchers and practitioners lack methods and empirical insights to make informed decisions about when and what types of human-like behaviors LLMs should exhibit. To fill this gap, we present a multi-dimensional analysis of the prevalence, potential effects, and controllability of these behaviors using LLM-as-a-judge and human evaluation. Across 21,000 multi-turn conversations from four widely used models (gpt-4o, gpt-4.1-mini, claude-sonnet-4.6, gemini-2.5-flash), we find that human-like behaviors are pervasive but vary across models and user factors (conversation goals and user profiles). In terms of perceived appropriateness, human evaluators judged self-referential and relationship-building behaviors as less appropriate from LLMs than from humans, but boundary-maintaining behaviors more appropriate from LLMs than from humans. Finally, we show that system prompting can control these behaviors, though it requires careful evaluation to avoid unintended effects. We discuss the implications of our findings and provide recommendations for responsible LLM design and evaluation.

Understanding the topology of decision regions is central to explaining the inner workings of deep neural networks. Prior empirical work has provided evidence that these regions are path connected. We study a stronger topological question: whether closed loops inside a decision region can be contracted without leaving that region. To this end, we propose an iterative quad-mesh filling procedure that constructs a finite-resolution label-preserving surface bounded by a given loop and lying entirely within the same decision region. We further connect this construction to natural Coons patches in order to quantify its deviation from a canonical geometric interpolation of the loop. By evaluating our method across several modern image-classification models, we provide empirical evidence supporting the hypothesis that decision regions in deep neural networks are not only path connected, but also simply connected.

Large language models (LLMs) are now largely involved in software development workflows, and the code they generate routinely includes third-party library (TPL) imports annotated with specific version identifiers. These version choices can carry security and compatibility risks, yet they have not been systematically studied. We present the first large-scale measurement study of version-level risk in LLM-generated Python code, evaluating 10 LLMs on PinTrace, a curated benchmark of 1,000 Stack Overflow programming tasks. LLMs tend to specify version identifiers when directly prompted at 26.83%-95.18%, while down to 6.45%-59.19% in creating a manifest file directly. Among the specified versions, 36.70%-55.70% of tasks contain at least one known CVE, and 62.75%-74.51% of them carry Critical or High severity ratings. In 72.27%-91.37% of cases, the associated CVEs were publicly disclosed before the model's knowledge cutoff. The statistics show all models converge on the same small set of risky release versions, indicating a systemic bias rather than isolated model error. Static compatibility rates range from 19.70% to 63.20%, with installation failure as the dominant cause. The dynamic test cases confirm the pattern by 6.49%-48.62% pass rates. Further experiments confirm that these failures are attributable to version selection rather than code quality, and that externally anchored version constraints substantially reduce both vulnerability exposure and compatibility failures. Our findings reveal LLM version selection as a first-class, previously overlooked risk surface in LLM-based development. We disclosed these findings to the community of the evaluated models, and several confirmed the issue. All the code and dataset have been released for open science at https://github.com/dw763j/PinTrace.

Synthetic data has transformed language model training, yet its role in time series forecasting remains poorly understood. We present a large-scale empirical study: nine experiment groups, 4,218 runs systematically evaluating synthetic time series augmentation across five architectures, four synthetic signals and seven datasets. The effect is sharply architecture-conditional: channel-mixing models (TimesNet, iTransformer) benefit in the majority of trials, while channel-independent models (DLinear, PatchTST) are consistently degraded. In selected low-resource settings the gains are striking: TimesNet trained on only 10% of Weather data with synthetic augmentation surpasses the full-data baseline (4 of 16 sparsity-dataset combinations). Averaged across all architectures, augmentation hurts in 67% of trials. We further find that only the Seasonal-Trend generator reliably helps across the tested benchmarks, and that hard curriculum switching is actively harmful (+24% MSE degradation). These results provide concrete, actionable guidelines on how to use synthetic data: use synthetic augmentation with channel-mixing architectures, use gradual annealing schedules, and treat low-resource augmentation as architecture- and dataset-dependent. Code is available at \href{https://github.com/hugoiscracked/synthetic-ts/tree/main}

Large language model (LLM)-based coding assistants have made substantial progress, yet most systems remain reactive, requiring developers to explicitly formulate their needs. Proactive coding assistants aim to infer latent developer intent from integrated development environment (IDE) interactions and repository context, thereby reducing interaction overhead and supporting more seamless assistance. However, research in this direction is limited by the scarcity of large-scale real-world developer behavior data. Existing studies therefore often rely on LLM-simulated IDE traces, whose fidelity to real development behavior remains unclear. In this paper, we investigate this simulation-to-reality gap through a large-scale empirical study. We collect real IDE interaction traces from 1{,}246 experienced industry developers over three consecutive days using a custom Visual Studio Code extension, and construct paired LLM-simulated traces for controlled comparison. Our analysis shows that simulated traces differ substantially from real traces in behavioral diversity, temporal structure, and exploratory patterns. Based on the collected data, we introduce \textbf{ProCodeBench}, a real-world benchmark for proactive intent prediction. Experiments with representative LLMs, retrieval-augmented methods, and agentic baselines show that current approaches remain far from reliable under real IDE traces, suggesting that simulation-based evaluation can overestimate real-world performance. Finally, our training study shows that simulated data cannot replace real data, but can complement it when used before real-world fine-tuning. These findings highlight the importance of real developer behavior data for evaluating and training proactive coding assistants.

As Vision-Language Models (VLMs) are increasingly deployed as autonomous cognitive cores for embodied assistants, evaluating their privacy awareness in physical environments becomes critical. Unlike digital chatbots, these agents operate in intimate spaces, such as homes and hospitals, where they possess the physical agency to observe and manipulate privacy-sensitive information and artifacts. However, current benchmarks remain limited to unimodal, text-based representations that cannot capture the demands of real-world settings. To bridge this gap, we present ImmersedPrivacy, an interactive audio-visual evaluation framework that simulates realistic physical environments using a Unity-based simulator. ImmersedPrivacy evaluates physically grounded privacy awareness across three progressive tiers that test a model's ability to identify sensitive items in cluttered scenes, adapt to shifting social contexts, and resolve conflicts between explicit commands and inferred privacy constraints. Our evaluation of 12 state-of-the-art models reveals consistent deficits. In cluttered scenes, all models exhibit monotonic performance decay as scene complexity grows due to perceptual deficit. When social context shifts, no model exceed 65% selection accuracy. Under conflicting commands, the best model gemini-3.1-pro perfectly balances task completion and privacy preservation in only 51% of cases. These findings reveal that current VLMs in the physical world suffer from perceptual fragility and fail to let their knowledge of privacy cues govern their situated behavior. Our code and data is available at https://github.com/immersed-privacy/immersed-privacy .

We report a systematic failure mode in predictive representation learning. Across 2695 neural network configurations trained to predict linear-Gaussian dynamics, the optimal encoder tracks the environment rather than the system it is meant to model. The mean causal fidelity -- the fraction of encoder sensitivity allocated to system degrees of freedom -- is 0.49, and only 2.5% of configurations exceed 0.70. The failure intensifies with dimension: at N=100, the optimal encoder becomes causally blind (fidelity ~10^{-8}) while achieving 92% lower prediction error than the causal representation. We prove this is not an optimization artifact but a structural property of the predictive objective: when environment modes are slower or less noisy than system modes, every minimizer of the population risk encodes the former. The set of dynamics exhibiting this predictive-causal gap is open and of positive measure in parameter space. In a nonlinear Duffing-GRU sweep, unconstrained predictors learn environment-dominant representations in 55% of tasks (95% CI 41--68%) versus 24% under operational grounding (p=2.3e-3); the median out-of-distribution MSE inflation under environment shift is 1.82x versus 1.00x. Operational grounding -- restricting the loss to system observables -- partially suppresses the gap, but causal fidelity is never recovered without an explicit system-environment boundary. The results identify the predictive-causal gap as a structural limit of learning, with implications for self-supervised representation learning, world models, and the scaling paradigm.

This revision updates a pop-to-jazz chord-generation rehearsal study. Best-epoch metrics still show that modest pop rehearsal preserves pop accuracy while improving jazz prediction, but v2 corrects released-checkpoint selection: the released F1 equals Phase 0, F2 had a transcription error, and ft-pop80-v2 restores a hash-distinct jazz-adapted F1 across 3 seeds.

As LLMs enter conflict monitoring, understanding systematic distortions in their outputs is critical for humanitarian accountability. We evaluate four vanilla open-weight models Gemma 3 4B, Llama 3.2 3B, Mistral 7B, and OLMo 2 7B and two domain-adapted models, AfroConfliBERT and AfroConfliLLAMA, on Nigeria and Cameroon conflict-event classification against ACLED, a gold-standard dataset with multi-stage verification. We find a bifurcated divergence in normative directionality. Open-weight models exhibit statistically significant False Illegitimation bias: Gemma misclassifies to 18.29% of legitimate battles as civilian-targeted violence while making zero False Legitimation errors. By contrast, AfroConfliBERT and AfroConfliLLAMA achieve near-directional neutrality, with Legitimization Bias differences indistinguishable from zero. Yet domain adaptation does not eliminate actor-based selection bias. Both adapted models show statistically significant actor bias comparable to vanilla LLMs; in Nigeria, state actors are legitimized 36.5% more often than non-state actors in identical tactical contexts. Open-weight outputs are also fragile to geography-specific lexical framing: delegitimizing phrases produce flip rates up to 66.7% in Cameroon and 34.2% in Nigeria, while perturbations salient in one context may not matter in another. Error trace profiling shows models mask normative bias through unfaithful rationale confabulations. In contrast, AfroConfliBERT and AfroConfliLLAMA are largely robust, with near-zero flip rates across perturbation categories. Overall, current models are not ready for unsupervised deployment in conflict monitoring. We call for fairness-aware fine-tuning to reduce actor-based selection bias, mandatory adversarial robustness evaluation against lexical manipulation, and context-specific human-in-the-loop oversight calibrated to regional difficulty.

Large language models (LLMs) employ safety mechanisms to prevent harmful outputs, yet these defenses primarily rely on semantic pattern matching. We show that encoding harmful prompts as coherent mathematical problems -- using formalisms such as set theory, formal logic, and quantum mechanics -- bypasses these filters at high rates, achieving 46%--56% average attack success across eight target models and two established benchmarks. Crucially, the effectiveness depends not on mathematical notation itself, but on whether a helper LLM deeply reformulates the harmful content into a genuine mathematical problem: rule-based encodings that apply mathematical formatting without such reformulation perform no better than unencoded baselines. We introduce a novel Formal Logic encoding that achieves attack success comparable to Set Theory, demonstrating that this vulnerability generalizes across mathematical formalisms. Additional experiments with repeat post-processing confirm that these attacks are robust to simple prompt augmentation. Notably, newer models (GPT-5, GPT-5-Mini) show substantially greater robustness than older models, though they remain vulnerable. Our findings highlight fundamental gaps in current safety frameworks and motivate defenses that reason about mathematical structure rather than surface-level semantics.

Neural operators have emerged as a powerful tool for learning mappings between infinite-dimensional function spaces. However, their approximation properties in Sobolev norms remain poorly quantified, even though these norms control both function values and derivatives and are the natural metrics for PDE well-posedness, stability, and generalization. We develop a functional-analytic framework for operator learning in Sobolev spaces and connect it to the numerical behavior of Fourier Neural Operators (FNOs) on a prototypical PDE. First, for a continuous nonlinear operator $\mathcal{G}: H^{s}(D)\to H^{t}(D')$ with $s > d/2$ and inputs restricted to a compact subset of $H^{s}(D)$, we prove that $\mathcal{G}$ can be uniformly approximated in $H^{t}$-norm by a neural operator with $\mathcal{O}(\varepsilon^{-d/s})$ trainable parameters. This yields an explicit complexity--error relation of the form $\|\mathcal{G}-\mathcal{G}_θ\|_{H^{t}} \lesssim C N^{-s/d}$. We then study the one-dimensional viscous Burgers solution operator $\mathcal{G}: u_{0}\mapsto u(\cdot,1)$ on a bounded $H^{1}$-ball and train FNOs with an $H^{1}$-loss. Across a sweep of model sizes, we obtain test $H^{1}$-errors down to $\mathcal{O}(10^{-7})$ and relative errors of order $10^{-3}$, with predictions accurately matching both solutions and spatial derivatives on held-out data. A log-log plot of Sobolev error versus parameter count exhibits an approximate power law $\|\mathcal{G}-\mathcal{G}_θ\|_{H^{1}} \approx C N^{-α}$ with empirical exponent $α\approx 1.4$, and long-horizon training reveals optimization instabilities in large FNOs, providing quantitative evidence that Sobolev-space approximation theory meaningfully predicts neural-operator scaling behavior.