Academic Intelligence · Curated Daily

探索全球前沿学术脉络

AcademicHub 汇聚顶级期刊与预印本平台的实时文献。定制您的专属科研雷达,利用大语言模型自动生成交叉领域文献分析简报。

01.
arXiv (CS.CV) 2026-06-12

VLADriveBench: Evaluating CoT-Action Relationship in VLA for Autonomous Driving

Vision-language-action (VLA) models generate chain-of-thought (CoT) reasoning alongside driving trajectories, but existing benchmarks evaluate only trajectory quality and do not assess whether the CoT is relevant, consistent, or causally connected to the driving action. We introduce VLADriveBench, a framework that combines observational metrics (mentioning, hallucination, contradiction, action alignment) with a CoT intervention protocol to provide complementary views of the CoT-action relationship. Applying VLADriveBench to three models across two architectures, we find that the two analyses can diverge sharply: ORION scores highest on observational alignment yet its CoT is epiphenomenal, while Alpamayo v1.5 scores lower yet its CoT is strongly causal, with visual salience gating the extent of CoT influence.

02.
arXiv (CS.AI) 2026-06-12

A Theory of Training Profit-Optimal LLMs

arXiv:2605.16430v3 Announce Type: replace-cross Abstract: Scaling LLMs requires tremendous computational resources, and recent advances in AI have gone hand in hand with massive amounts of capital expenditure. While it is established that scaling up LLMs reliably increases model quality (quantified in terms of loss or downstream evaluations), it is unclear how these quality improvements translate to potential revenue, and whether revenue increases would offset costs of larger-scale training and inference. In this work, we develop an economic model for characterizing the rational behavior of an LLM training firm by combining scaling laws with microeconomic theory. Under our model of firm behavior, LLM quality can be increased with more parameters and training tokens, leading to more potential adoption by consumers, who each have a quality threshold for using the LLM. On the other hand, additional parameters and training tokens both incur additional costs. We analyze the profit maximization problem for this model under compute-bound and data-bound regimes. In the compute-bound regime, optimal model size and token budget track hardware efficiency $E$ (FLOPs/\$) at a near-linear rate; total training cost then scales sub-quadratically in $E$. Data efficiency improvements incentivize larger models and training expenditure. When we are limited to $D$ data, profit-optimal training expenditure scales as $D^2/E$, i.e, increase with data and decreases with hardware efficiency (as well as data efficiency). Finally, we analyze practical trends in training expenditure: current trends are consistent with our most permissive model variants in the compute-bound regime, but are not profit-optimal in the data-bound regime or assuming hardware advances will stall. Overall, our results provide a theory of profit-optimal LLM training, providing a foundation for engaging critically with industry statements and supporting long-term economic decision making.

03.
arXiv (CS.CV) 2026-06-17

Revisiting Structural Dependency in Autoregressive Multi-Task Table Recognition via Order-Independent Cell-Level Representations

Multi-task table recognition jointly addresses table structure prediction, cell localization, and cell content recognition within a unified framework. Existing approaches often rely on autoregressive decoders to generate table structures and reuse their hidden states for cell localization and content recognition. This autoregressive generation process can make cell representations order-dependent, degrading global consistency across cells. This paper proposes a structural refinement module that produces order-independent cell features through non-causal attention. This design enables parallel inference of cell contents while conditioning each cell on global context encoded in the refined features. Experiments on two large datasets demonstrate consistent gains in cell localization and end-to-end recognition, while reducing overall inference time by around threefold.

04.
medRxiv (Medicine) 2026-06-15

Unveiling the Awareness of Private Health Insurance Coverage among Healthcare Professionals in Freetown, Sierra Leone: Insights Extracted from Their Perspectives.

Our study is an assessment of the knowledge, personal coverage, and related determinants of private health insurance as revealed by healthcare professionals in Freetown, the urban capital of Sierra Leone. This study stands as a precursor for Low- and Middle-Income Countries (LMICs), like Sierra Leone, seeking to establish Universal Health Coverage (UHC) to provide healthcare access and coverage through publicly arranged risk pooling, designed to help protect against unmanageable medical costs. In parallel, such countries face significant challenges with achieving sustainable universal coverage due to limited public resources, inefficient allocation systems, uneasy reliance on out-of-pocket payments, and large struggling populations. Our research sheds particular light on how healthcare professionals view their own participation with private healthcare options. A cross-sectional, analytical study was conducted, openly recruiting individuals from various facilities in Freetown. Using the Yamane Formula, a sample size of 109 participants was calculated. STATA 14.0 was used for data analysis. Our findings revealed that 96 (88.9%) participants did not have private health insurance, while 12 (11.1%) did have private coverage. However, 105 (97.2%) reported other modes of health insurance, with only 3 (2.8%) uninsured. Notably, 97.2% expressed willingness to join a private health insurance scheme. Our study found no statistically significant associations between selected indicators (demographic or socioeconomic fac tors) and current insurance coverage among study participants. These results highlight a low prevalence and understanding of private health insurance among healthcare professionals in a representative urban center in Sub-Saharan Africa (SSA), while acknowledging high willingness to enroll. The lack of any significant determinants suggests other unexamined factors, such as cost, accessibility, or awareness, capable of influencing the adoption and implementation of a universal health program.

05.
arXiv (CS.AI) 2026-06-15

Learning What to Predict: Downstream-Guided Task Design for Continued Pretraining

arXiv:2601.22108v2 Announce Type: replace-cross Abstract: Continued pretraining is optimized with fixed self-supervised tasks but selected by downstream performance, creating a coarse feedback loop in which practitioners evaluate checkpoints, change data mixtures or objectives, and restart runs, while individual updates remain blind to target capabilities. We ask whether a small set of verifiable downstream examples can provide step-level feedback without directly supervising the learner. We introduce V-pretraining, which decouples a learner trained only with a self-supervised loss from a lightweight task designer that constructs targets or views for unlabeled batches. Given the current learner and batch, V-pretraining scores a candidate construction by predicting the first-order reduction in downstream loss after the induced self-supervised update. The designer maximizes this value; the learner then applies the update with targets or views detached, so downstream labels never update learner parameters. We instantiate V-pretraining as adaptive top-K soft targets for language modeling and learned views or masks for self-supervised vision. Across both modalities, V-pretraining improves target capabilities without degrading generalization. Under wall-clock-matched continued pretraining, it improves GSM8K Pass@1 for Qwen models using 1,024 GSM8K examples only as feedback, including a +7.4 point single-run gain for Qwen2.5-0.5B. In vision, it improves DINOv3 transfer to ADE20K semantic segmentation and NYUv2 depth estimation while preserving ImageNet linear accuracy, suggesting that feedback-guided task construction can improve target capabilities without collapsing general-purpose representations.

06.
arXiv (CS.AI) 2026-06-12

Standardized Methods and Recommendations for Green Federated Learning

arXiv:2602.00343v2 Announce Type: replace-cross Abstract: Federated learning (FL) enables collaborative model training over privacy-sensitive, distributed data, but its environmental impact is difficult to compare across studies due to inconsistent measurement boundaries and heterogeneous reporting. We present a practical carbon-accounting methodology for FL CO2e tracking using NVIDIA NVFlare and CodeCarbon for explicit, phase-aware tasks (initialization, per-round training, evaluation, and idle/coordination). To capture non-compute effects, we additionally estimate communication emissions from transmitted model-update sizes under a network-configurable energy model. We validate the proposed approach on two representative workloads: CIFAR-10 image classification and retinal optic disk segmentation. In CIFAR-10, controlled client-efficiency scenarios show that system-level slowdowns and coordination effects can contribute meaningfully to carbon footprint under an otherwise fixed FL protocol, increasing total CO2e by 8.34x (medium) and 21.73x (low) relative to the high-efficiency baseline. In retinal segmentation, swapping GPU tiers (H100 vs.\ V100) yields a consistent 1.7x runtime gap (290 vs. 503 minutes) while producing non-uniform changes in total energy and CO2e across sites, underscoring the need for per-site and per-round reporting. Overall, our results support a standardized carbon accounting method that acts as a prerequisite for reproducible 'green' FL evaluation. Our code is available at https://github.com/Pediatric-Accelerated-Intelligence-Lab/carbon_footprint.

07.
arXiv (CS.AI) 2026-06-15

TwinBI: An Agentic Digital Twin for Efficient Augmented Interactions with Business Intelligence Dashboards

arXiv:2606.13731v1 Announce Type: new Abstract: Business intelligence (BI) increasingly combines dashboard interaction with LLM-based assistance, but these two modes often fall out of sync during multi-step analysis. As users switch between direct dashboard manipulation and natural-language queries, it becomes difficult to preserve a consistent analytical state across filters, hierarchies, metrics, and chart context. We present TwinBI, an agentic digital-twin framework that couples an LLM-based agent system with an executable BI dashboard state. TwinBI unifies conversational interaction, dashboard manipulation, semantic grounding, and provenance tracking through a shared analytical state reconstructed from a unified interaction log. It also exposes artifacts such as schema views, SQL, logs, and an /insights command for state-grounded analytical summaries. We evaluate TwinBI in two complementary ways. In a controlled A/B benchmark with the same backbone agent, TwinBI improves exact-match accuracy from 43.3% to 63.3%, partial-credit accuracy from 48.3% to 70.8%, and substantially reduces timeout rate from 40.0% to 10.0% relative to Dashboard alone. In a usability study, participants benefited from the integrated dashboard-and-chat workflow, with high task accuracy, moderate workload, and favorable ratings for state-aware interaction mechanisms. These results suggest that TwinBI improves both agent-level analytical reliability and user-facing analytical support by turning visible dashboard state into richer actionable context. Our dataset and source code are available at: https://github.com/simonjisu/TwinBI

08.
PLOS Medicine 2026-05-27

Sequential chemo-immunotherapy followed by standard versus reduced thoracic radiotherapy for older and/or frail stage III non-small-cell lung cancer: A randomized open-label cohort trial

作者:

by Wei-Xiang Qi, Shuyan Li, Mengdi Wang, Huan Li, Feifei Xu, Lei Yao, Biao Yu, Linlin Chen, Gang Cai, Cheng Xu, Xianwen Sun, Zhiyao Bao, Jiayi Chen, Yi Xiang, Shengguang Zhao Background The appropriateness of concurrent chemoradiotherapy (cCRT) for older or clinically vulnerable stage III unresectable non-small-cell lung cancer (NSCLC) patients remains contentious. Furthermore, the survival implications of de-escalating thoracic radiotherapy (RT) intensity in this population have not been conclusively elucidated. Methods and findings We conducted a phase II randomized, open-label, two-cohort (non-comparative) trial at a tertiary hospital in China (NCT05557552). Between September 30, 2022 and April 30, 2024, we enrolled 56 older and/or frail patients with stage III NSCLC who were ineligible for cCRT. The primary endpoint was the 1-year progression-free survival (PFS) rate estimated using the Kaplan–Meier method. Secondary endpoints included objective response rate (ORR), overall survival (OS), and safety. In the intention-to-treat (ITT) set, which included all 56 randomized patients who received at least one dose of study treatment, the 1-year PFS was 84.3% (95% confidence interval [CI] [70.3%, 98.3%]) in the standard RT group and 70.7% (95% CI [54.3%, 87.1%]) in the reduced RT group. In the per-protocol set (53 patients), the 1-year PFS was 82.9% (95% CI [68.9%, 98.8%]) in the standard RT group and 73.4% (95% CI [58.3%, 92.4%]), with a median follow-up of 24 months. Among 56 patients in the safety analysis set, 71.4% of patients experienced grade 3/4 adverse events (AEs) in the standard RT group and 53.6% in the reduced RT group. One patient (3.6%) in the reduced RT and three patients (10.7%) in the standardized RT experienced grade 5 AEs. The main limitations are the non-comparative design, small sample size, and lack of power to establish non-inferiority or superiority. Conclusion The current study suggested that reduced RT combined with sequential chemo-immunotherapy might be feasible for older/frail patients intolerant to cCRT, showing numerically similar survival outcomes. These exploratory findings warrant confirmation in larger, adequately powered randomized trials. Trial registration The trial had been registered on ClinicalTrials.gov on Sep 30, 2022.ClinicalTrials.gov NCT05557552

09.
arXiv (CS.CL) 2026-06-15

Benchmarking Web Agent Safety under E-commerce Deceptive Interfaces

As autonomous web agents are increasingly deployed to perform real-world tasks, ensuring their safety has become a critical concern. In this work, we study web agent behavior under realistic deceptive interfaces in the e-commerce domain. We introduce WebDecept, a lightweight and configurable plugin framework that enables controlled injection of deceptive interface patterns into existing web environments. Using WebDecept, we instantiate seven deceptive patterns commonly observed on the open web, including targeted advertisements, domain redirection, and shopping manipulation. By injecting these patterns into the frontend during task execution, we perform controlled evaluation of multiple multimodal web agents. Our results show that current web agents are highly susceptible to multiple classes of deceptive interfaces, and that prompt-based constraints are often insufficient to mitigate these failures. We further analyze how the design choices of deceptive patterns influence the success of such manipulations. These findings highlight safety challenges that should be addressed as web agents are scaled toward real-world deployment.

10.
arXiv (CS.AI) 2026-06-11

Noise-Guided Transport for Imitation Learning

arXiv:2509.26294v2 Announce Type: replace-cross Abstract: We consider imitation learning in the low-data regime, where only a limited number of expert demonstrations are available. In this setting, methods that rely on large-scale pretraining or high-capacity architectures can be difficult to apply, and efficiency with respect to demonstration data becomes critical. We introduce Noise-Guided Transport (NGT), a lightweight off-policy method that casts imitation as an optimal transport problem solved via adversarial training. NGT requires no pretraining or specialized architectures, incorporates uncertainty estimation by design, and is easy to implement and tune. Despite its simplicity, NGT achieves strong performance on challenging continuous control tasks, including high-dimensional Humanoid tasks, under ultra-low data regimes with as few as 20 transitions.

11.
arXiv (CS.LG) 2026-06-19

The Hidden Environmental Cost of Poor Coding Practices in TensorFlow and Keras Applications: A Study on Resource Leaks and Carbon Emissions

arXiv:2606.19799v1 Announce Type: cross Abstract: Efficiency and sustainability are critical considerations in the development and deployment of machine learning (ML) applications. Among the factors influencing sustainability, resource leaks in ML code can introduce hidden inefficiencies that elevate energy consumption and CO2 emissions. Despite this, empirical evidence quantifying their environmental impact remains limited. This emerging results paper presents an initial empirical investigation of two common resource-leak smells, namely Improper Model Reuse (IMR) and Unreleased Tensor References (UTR), and their impact on energy consumption and CO2 emissions in TensorFlow and Keras workloads. Controlled experiments were conducted for each smell by executing identical training tasks while comparing against a smell-free baseline. Our preliminary results show that both smells consistently increase estimated electricity usage and carbon emissions. IMR and UTR increased electricity consumption by approximately 32% and 46%, respectively, with proportional increases in CO2 emissions. Paired statistical tests indicate that these differences are systematic and statistically significant, providing initial empirical evidence that resource-leak smells may degrade ML energy efficiency and environmental sustainability. These findings suggest that resource-leak smells pose measurable risks to both software quality and sustainability, emphasizing the importance of integrating resource-lifecycle management and energy-efficiency considerations into ML development.

12.
arXiv (CS.CV) 2026-06-12

RepWAM: World Action Modeling with Representation Visual-Action Tokenizers

This work presents RepWAM, a representation-centric world action model (WAM) built on representation visual-action tokenizers. Existing WAMs typically inherit reconstruction-oriented video tokenizers from pretrained video generation models. Although these tokenizers preserve visual fidelity, pixel reconstruction alone provides limited guidance for learning instruction-following dynamics that connect future prediction with robot control. To address this, we explore a semantic visual-action latent space for representation-centric world action modeling. Specifically, we train a representation visual-action tokenizer that maps visual inputs into aligned visual and latent action tokens. We then pretrain our WAM to jointly model future visual states and the latent actions that connect them under language instructions, followed by adaptation to real robot trajectories for closed-loop manipulation. Experiments on real-world manipulation tasks and simulation benchmarks show that RepWAM delivers strong performance across diverse manipulation settings, while ablations highlight the value of semantic visual-action tokenization over reconstruction-oriented alternatives. These results establish representation visual-action tokenization as a promising foundation for world action models and a step toward generalist robot policies. Code and weights will be available at https://github.com/wdrink/RepWAM.

13.
medRxiv (Medicine) 2026-06-15

Shortened blastocyst vitrification achieves live birth rates comparable to standard protocols: an analysis of 3168 cryotransfers

Study question Do shortened blastocyst vitrification and warming protocols provide comparable live birth rates (LBR) and obstetrical and perinatal outcomes to traditional vitrification and warming protocols? Summary answer Shortened vitrification and warming protocols provide comparable LBR, obstetric and perinatal outcomes to traditional protocols. Shortened vitrification coupled with traditional multi step warming benefitted women >35yrs. What is known already Embryo viability following cryopreservation is dependent on blastomere survival and functional integrity, both impacted by ice crystal formation and osmotic gradients. Recent innovations in cryopreservation challenge the need for stepwise dehydration and rehydration protocols. While one step ''fast'' blastocyst warming protocols seem to provide equivalent clinical outcomes to traditional ''slow'' protocols, fewer studies investigate whether blastocyst dehydration rates can be similarly increased. A thorough safety and effectiveness evaluation remains necessary for both treatment success and offspring health. Study design, size, duration Three clinics within a network participated in this retrospective consecutive cohort study, with cycle data collected for 3603 warmed blastocysts resulting in 3168 frozen blastocyst transfers in 2170 patients between 2023 and 2025. We modelled the relationship between ''fast'' versus ''slow'' protocols and outcomes with Generalized Additive Models, and linear and logistic regressions where appropriate. Two tailed chi square with Yates correction was used to examine pregnancy loss and obstetrical and perinatal outcomes; p0.05). Importantly, women 35yrs or older at vitrification (n=1715 transfers) profited from a F/S strategy, which provided a significant increase in live birth rates (OR:1.42 [1.02-1.98] p=0.038) compared to S/S. The same improved live birth following a F/S strategy were also seen in embryos of lower quality (OR:1.78 [1.12-2.83] p=0.015), suggesting of a protective effect of this cryopreservation strategy on the developmental competence of impaired germplasm. Limitations, reasons for caution Factors affecting the results may be unaccounted for by the study retrospective nature. Wider implication of the findings Overall, shortened, ''faster'' vitrification and warming protocols provide comparable reproductive outcomes to traditional ones. The combination of shorter exposure to cryoprotectant (CPA) during vitrification and stepwise osmotic gradient during warming provided significant clinical benefits specifically to patients >35 and lower quality embryos, pointing to the possibility of adapting vitrification protocols to specific patients populations and optimizing their clinical outcomes.

15.
arXiv (CS.LG) 2026-06-12

Adaptive generative moment matching networks for improved learning of dependence structures

arXiv:2508.21531v2 Announce Type: replace-cross Abstract: An adaptive bandwidth selection procedure for the mixture kernel in the maximum mean discrepancy (MMD) for fitting generative moment matching networks (GMMNs) is introduced, and improved learning of copula random number generators is demonstrated. Based on the relative error of the training loss, the number of kernels is increased during training; additionally, the relative error of the validation loss is used as an early stopping criterion. While training time remains similar, adaptively training GMMNs (AGMMNs) significantly increases training performance, which is shown based on validation MMD trajectories, samples and validation MMD values. Superiority of AGMMNs over GMMNs and parametric copula models is also demonstrated in terms of three applications. First, convergence rates of estimators based on quasi-random versus pseudo-random samples from copulas are investigated in dimensions as large as 100 for the first time. Second, replicated validation MMDs, as well as Monte Carlo and quasi-Monte Carlo applications demonstrate the improved training of AGMMNs for a copula model implied by the 50 constituents of the S&P 500 index after deGARCHing. Last, both the latter dataset and 50 constituents of the FTSE 100 are used to demonstrate that the improved training of AGMMNs indeed translates to an improved model prediction.

16.
arXiv (CS.CV) 2026-06-17

Bounding Box Label Propagation for Re-Annotation of Document Layout Analysis Datasets

Datasets in practical document processing scenarios typically grow over time, and their class annotations undergo continuous refinement. This creates significant re-annotation efforts, which are time-consuming and costly. A promising remedy is to re-annotate only a small subset of available documents manually and apply semi-supervised learning techniques that leverage both labelled and unlabelled data. Although there are numerous approaches to tackle this problem for classification, there exists no adaptation for the problem of re-classifying object detection instances, e.g. for document layout analysis. To this end, we propose Bounding Box Label Propagation (BBLP), a pseudo-labelling framework for object detection. An object encoder integrates visual, textual, and positional embeddings from object detection samples to come up with a joint embedding that can be used for Label Propagation on partially annotated datasets in a plug-and-play fashion. Evaluation results indicate that the proposed approach produces high-quality class annotations of bounding boxes. In the D4LA layout analysis dataset, it achieves a mAP of 54.0%, corresponding to 81.6% of fully supervised performance, while using only 10% labelled data. Our work demonstrates the potential of Label Propagation for object detection and lays the groundwork for reducing manual annotation efforts in real-world document processing applications.

17.
bioRxiv (Bioinfo) 2026-06-14

Generative design of antigen-specific T-cell receptor sequences with a conditional diffusion model

T cell receptor (TCR)-based immunotherapy holds immense potential for treating cancers and infectious diseases, where highly antigen-specific TCR recognition is crucial for adaptive immunity against tumors and pathogens. Engineering or de novo generation of the complementarity-determining region 3 (CDR3) loops of TCRs using artificial intelligence offers a powerful alternative to designing reactive TCRs rather than laborious experimental screening. However, current in silico approaches are constrained by weak conditional guidance, limited flexibility, and a lack of rigorous functional validation. To address these limitations, we introduce TCRDiff, a generative diffusion framework for designing antigen-specific TCRs conditioned on peptide-MHC (pMHC) targets and germline-encoded variable genes. By leveraging pre-trained knowledge from massive T-cell repertoires and TCR-pMHC recognition data, TCRDiff generates CDR3{beta} sequences with state-of-the-art fidelity to native binding TCRs through a denoising diffusion process. Furthermore, incorporating the interface geometry features generated TCR-pMHC complexes with superior structural plausibility. As a proof of concept, we deployed TCRDiff in a systematic pipeline to design candidate TCRs for immunotherapy. In vitro activation assays validated that TCRDiff-generated TCRs specifically recognize the MAGE-A3 epitope with minimized off-target cross-reactivity. Together, TCRDiff establishes a powerful, validated computational paradigm to accelerate the development of TCR-based immunotherapies.

18.
arXiv (CS.CL) 2026-06-17

Algorithmic Prompt Generation for Diverse Human-like Teaming and Communication with Large Language Models

Understanding how humans collaborate and communicate in teams is essential for improving human-agent teaming and AI-assisted decision-making. However, relying solely on data from large-scale user studies is impractical due to logistical, ethical, and practical constraints, necessitating synthetic models of multiple diverse human behaviors. Recently, agents powered by Large Language Models (LLMs) have been shown to emulate human-like behavior in social settings. But, obtaining a large set of diverse behaviors requires manual effort in the form of designing prompts. On the other hand, Quality Diversity (QD) optimization has been shown to be capable of generating diverse Reinforcement Learning (RL) agent behavior. In this work, we combine QD optimization with LLM-powered agents to iteratively search for prompts that generate diverse team behavior in a long-horizon, multi-step collaborative environment. We first show, through a human-subjects experiment, that humans exhibit diverse coordination and communication behavior in this domain. We then present a series of experiments showing that our approach captures behaviors that are difficult to observe without large-scale data collection, and a follow-up user study to show that these generated behaviors are human-like. Our findings highlight the combination of QD and LLM-powered agents as an effective tool for studying teaming and communication strategies in multi-agent collaboration.

19.
arXiv (CS.AI) 2026-06-19

Learner-based Concept Drift Detection: Analysis and Evaluation

arXiv:2606.20216v1 Announce Type: cross Abstract: Machine learning algorithms deployed for evolving streaming environments must handle the non-stationary data distributions, commonly referred to as concept drift. The presence of concept drift poses a major challenge for many real-world applications because it can severely degrade their predictive performance, hindering their ability to support robust decision-making. Consequently, the timely and efficient detection of drift events is critical for sustaining high accuracy over time. This study examines theoretically the concept drift characteristics and numerous drift detection algorithms across several categories. Furthermore, we evaluate their performance on both synthetic and real-world datasets exhibiting diverse streaming scenarios and drift characteristics, such as abrupt and gradual changes. This study aims to enhance understanding of the complex notion of concept drift characteristics and behavior of drift detectors, along with their applicability to diverse contexts.

20.
arXiv (CS.CL) 2026-06-17

Structural Role Injection in Handlebars-Templated LLM Prompts: Triple-Brace Interpolation, Delimiter Family, and the Limits of HTML Auto-Escaping

Large language model applications build prompts from templates, and Handlebars is a widely used templating engine and the default prompt-template format in Microsoft Semantic Kernel. Its double-brace {{x}} expression HTML-escapes the interpolated value and is documented as the safe default; its triple-brace {{{x}}} expression inserts the value raw. We show that this choice silently governs an application's exposure to structural role injection, where attacker-controlled data carries chat role delimiters that forge a higher-privilege turn. A model-free analysis establishes the mechanism: Handlebars escaping rewrites angle brackets but not square brackets, colons, or Markdown hashes, so it neutralises ChatML, Llama-3, and XML role delimiters (survival rate 0.00) while leaving Llama-2 [INST], legacy Human:/Assistant:, and Markdown ### delimiters intact (survival rate 1.00 for the last two). We then run 5760 trials across seven delimiter families, two attack objectives, and four models (GPT-3.5 Turbo, GPT-4o mini, GPT-4.1 mini, Claude Haiku 4.5) at a combined API cost of 1.63 USD. GPT-3.5 Turbo follows the task-hijack instruction in 97% of raw and 91% of escaped trials, with the escaping protection concentrated in the angle-bracket families and absent for the colon- and Markdown-based families; the harder secret-exfiltration objective, which does not saturate, exposes the same family interaction more cleanly. Claude Haiku 4.5 resists both objectives almost entirely. The escaped default protects only the delimiter schemes whose characters HTML escaping happens to cover, gives no protection for the rest, and cannot substitute for a structural separation of instruction and data.

21.
arXiv (CS.CL) 2026-06-17

LVLMs and Humans Ground Differently in Referential Communication

For generative AI agents to partner effectively with human users, the ability to accurately predict human intent is critical. But this ability to collaborate remains limited by a critical deficit: an inability to model common ground. We present a referential communication experiment with a factorial design involving director-matcher pairs (human-human, human-AI, AI-human, and AI-AI) that interact with multiple turns in repeated rounds to match pictures of objects not associated with any obvious lexicalized labels. We show that LVLMs cannot interactively generate and resolve referring expressions in a way that enables smooth communication, a crucial skill that underlies human language use. We release our corpus of 356 dialogues (89 pairs over 4 rounds each) along with the online pipeline for data collection and the tools for analyzing accuracy, efficiency, and lexical overlap.

22.
arXiv (CS.CV) 2026-06-16

Beyond Scalar Rewards by Internalizing Reasoning into Score Distributions

Reward models are central to text-to-image post-training, but visual preference is subjective and better represented as a distribution over rubric scores than as a deterministic scalar. Existing scalar, score-token, and pairwise reward models over-compress uncertainty and fine-grained score differences, while reasoning-based generative rewards provide stronger judgments but are costly to deploy and difficult to use as direct optimization signals. We propose Z-Reward, a teacher-student reward modeling framework that decouples reasoning-heavy judgment from efficient reward deployment. The teacher is a large VLM that uses reasoning to infer rubric-aligned score distributions, and is trained with Group-wise Direct Score Optimization (GDSO), which combines policy-gradient rewards from distribution expectations with direct pointwise and pairwise supervision on score distributions and score gaps. The student is trained with Reasoning-Internalized Score Distillation (RISD), which transfers the teacher's reasoning-conditioned score distribution into a compact VLM without requiring explicit reasoning chains at inference time. On our internally annotated evaluation set, the 27B GDSO teacher reaches 89.6% human preference accuracy, outperforming SFT, RewardDance, and GRPO, while the 9B RISD student reaches 88.6%, outperforming the OPD baseline and closely matching the larger teacher. We further show that Z-Reward can serve as a differentiable reward signal for text-to-image optimization, yielding a 41.3% net human-preference improvement over the SFT baseline.

23.
arXiv (CS.AI) 2026-06-11

Sparse probes and murky physics: a case study of interpretability challenges in a foundation model for continuum dynamics

arXiv:2606.11657v1 Announce Type: cross Abstract: Generative AI emulators are increasingly used in scientific domains where we already have strong theory, benchmarks, and physical intuition. This raises a central evaluation and interpretability question: when a foundation-style model can reproduce known continuum dynamics, what internal mechanism supports that behavior, is the internal behaviour consistent with known physics, and how does it relate to where the emulator succeeds or fails? We investigate a cross-domain foundation model for continuum dynamics, Walrus by Polymathic, using mechanistic interpretability guided by physical principles. We apply a sparse autoencoder (SAE) to probe a selected layer, and address the practical challenge of triaging a large feature set (over 20,000) using enstrophy as a physically grounded metric. As a deliberately simple testbed, we focus on shear flow and compare feature recruitment across multiple shear-flow setups, i.e. parameter values in the numerical simulation. Across setups we find evidence of piecewise consistency, with subsets of features recurring in similar roles, but this structure is intermittent and does not map cleanly onto standard physical decompositions. In parallel, direct comparisons between numerical simulation and the emulator reveal systematic output-level discrepancies, including regimes where energy/structures become too diffuse or too localized. We connect parts of these discrepancies to changes in specific SAE feature usage. Our work highlights open questions for scientific foundation models: how to robustly prioritize mechanistically meaningful features, how to separate stable structure from analysis artifacts (including single-layer and SAE limitations), and how to use established benchmarks to decide when "different" internal representations are genuinely informative rather than merely effective.

24.
arXiv (CS.AI) 2026-06-17

Trust-Aware Multi-Agent Traceability: Confidence-Calibrated Knowledge Graphs for Consistent Software Artifact Management

arXiv:2606.17203v1 Announce Type: cross Abstract: Multi-agent AI systems are increasingly used to automate software engineering tasks including requirements analysis, architecture design, test generation, and traceability linking. When these agents operate as a sequential pipeline over shared software artifacts, errors and low-confidence decisions made by upstream agents propagate to downstream stages, producing orphaned requirements, contradictory links, and compliance gaps that pose significant risks in safety-critical domains. We propose a trust-aware coordination framework where a shared knowledge graph serves as both centralized semantic memory and a coordination surface through which agents assess and build upon each other's contributions using calibrated confidence scores. Our approach introduces a two-stage traceability link prediction pipeline combining embedding-based retrieval with LLM-based multi-criteria analysis, a traceability seeding mechanism that enables comparison between derivation-time and validation-time confidence, and a consistency protocol governing pipeline interactions through confidence threshold gating, confidence divergence detection, and conflict resolution. We evaluate on an automotive software engineering case study measuring link prediction calibration, protocol effectiveness, threshold sensitivity, and the impact of traceability seeding. Ablation studies confirm that confidence calibration is essential for effective pipeline coordination.

25.
arXiv (CS.LG) 2026-06-19

Evaluating Universal Machine Learning Force Fields Against Experimental Measurements

arXiv:2508.05762v2 Announce Type: replace-cross Abstract: Universal machine learning force fields (UMLFFs) promise to revolutionize materials science by enabling rapid atomistic simulations across the periodic table. However, their evaluation has been limited to computational benchmarks that may not reflect real-world performance. We introduce UniFFBench, a comprehensive evaluation framework featuring the MinX dataset – a diverse collection of 1,500+ mineral systems spanning 85 elements, extreme thermodynamic conditions (0–5000 K, 0–1000 GPa), and structural complexity, including partial occupancy and disorder. This diversity, combined with experimental reference values for validation, enables assessment of UMLFF generalization across chemical space and conditions substantially beyond typical training scenarios. Our systematic evaluation of six state-of-the-art UMLFFs reveals a substantial ``reality gap'': models achieving impressive performance on computational benchmarks often fail when confronted with experimental complexity. Even the best-performing models exhibit higher density prediction error than the threshold required for practical applications. We observe disconnects between simulation stability and mechanical property accuracy, with prediction errors correlating with training data representation rather than the modeling method.