artificial intelligence

Artificial intelligence–driven decision-making after endoscopic resection for early gastric cancer

Artificial intelligence framework for predicting inclined magneto-bioconvective Ellis penta-hybrid nanofluid flow over a stretching cylinder with gyrotactic microorganisms: A Levenberg–Marquardt approach

AI-Driven Sustainability in Industry 5.0: The Role of Responsible Leadership

Abstract As Industry 5.0 advances, artificial intelligence (AI) is increasingly positioned as a driver of sustainability in knowledge-based economies; however, empirical outcomes remain uneven and frequently symbolic. Addressing this paradox, this paper examines AI-driven sustainability through the lens of knowledge creation, governance, and application, rather than technological capability alone. Using a Critical Interpretive Synthesis, the paper systematically analyses interdisciplinary literature on AI, Industry 5.0, sustainability, and leadership to move beyond descriptive aggregation toward theory development. The findings reconceptualise AI as a knowledge infrastructure whose sustainability value depends on how AI-generated knowledge is governed, interpreted, and applied across systems. The paper further advances theory by reframing responsible leadership as a knowledge-governance mechanism, explaining how leadership shapes the prioritisation and diffusion of AI-enabled knowledge across micro (individual), meso (organisational/industry), and macro (institutional) levels. Building on these insights, the paper develops an integrative framework that explains why AI-enabled sustainability initiatives often result in performative environmental, social, and governance (ESG) compliance rather than substantive environmental and social impact. By linking responsible leadership with AI knowledge governance, the paper contributes to the knowledge-economy literature by explaining variability in sustainability outcomes beyond technological adoption. The paper concludes by outlining implications for organisational governance and identifying directions for future empirical research to test and extend the proposed framework.

AI-Powered Automated Speaking Scoring and Human Evaluation: A Comparative Study of EFL Learners’ Oral Proficiency

Recent advances in artificial intelligence have led to the development of automated speaking assessment systems capable of evaluating oral proficiency with increasing accuracy. This study compares Artificial Intelligence (AI) powered automated speaking scoring and human evaluation in assessing the oral proficiency of 74 Saudi English as Foreign Language (EFL) learners. Participants completed a picture-description speaking task, which was evaluated by both Claude and two trained human raters using identical holistic and analytic speaking rubrics. The study examined holistic speaking scores and five analytic dimensions: coherence, cohesion, content development, grammar, and vocabulary. Statistical analyses included descriptive statistics, comparative analyses, Intraclass Correlation Coefficients (ICC), Weighted Kappa coefficients, and Bland-Altman analysis. Results showed no significant difference between AI-generated and human-assigned holistic speaking scores. Similarly, cohesion and grammar demonstrated strong similarity between the two assessment approaches. However, significant differences were observed for content development, vocabulary, and coherence, with Claude consistently assigning slightly higher scores than the human raters. Agreement analyses revealed good-to-strong agreement across both holistic and analytic assessments. The findings suggest that AI-powered speaking assessment can produce evaluations broadly comparable to human judgments while demonstrating strong consistency across multiple dimensions of oral proficiency. The study supports the potential of AI-assisted speaking assessment as a reliable complement to human evaluation in EFL contexts.

Loss of Environmental Awareness in Businesses: Organizational Blindness

Organizations operating in increasingly dynamic and uncertain environments face growing challenges in recognizing and responding to external changes. This study examines the phenomenon of organizational blindness, defined as the systematic inability of organizations to perceive, interpret, and act upon critical environmental signals despite the availability of relevant information. Drawing on theories of organizational cognition, managerial attention, sensemaking, and strategic management, the study explores the cognitive, structural, and cultural mechanisms that contribute to this deficiency. It analyzes key concepts including bounded rationality, dominant logic, cognitive rigidity, organizational inertia, information-processing failures, organizational silence, and institutional isomorphism, demonstrating how these factors collectively restrict strategic adaptation. To illustrate the practical consequences of organizational blindness, the study examines the well-known cases of Kodak, Nokia, and Blockbuster, showing how established routines, overconfidence, and rigid mental models prevented these organizations from responding effectively to technological and market transformations. The findings suggest that organizational blindness results not from a lack of information but from failures in attention, interpretation, communication, and decision-making processes. To overcome these challenges, the study proposes several managerial strategies, including strengthening environmental scanning capabilities, promoting cognitive diversity within leadership teams, encouraging constructive dissent, improving cross-functional communication, and developing organizational ambidexterity that balances operational efficiency with innovation and exploration. The study concludes that organizations capable of detecting weak environmental signals and adapting proactively are better positioned to sustain competitive advantage in turbulent environments. It further recommends future research on the role of digital technologies, artificial intelligence, and real-time analytics in enhancing organizational awareness while also examining whether these technologies may create new forms of organizational blindness.

Deconstructing and Reconstructing University Classroom Authority in the Age of Generative AI: Practical Reflections on a Project Based Learning Reform

The rapid proliferation of generative artificial intelligence is fundamentally reshaping higher education, challenging the traditional lecture-based, knowledge-transmission model of classroom instruction. This paper offers a reflective analysis based on the author's first-hand teaching experience at a Chinese university with a finance and economics focus, where two AI-related courses are offered: a general-education AI literacy course for all undergraduates and an advanced deep learning course for computer science majors. The analysis reveals that AI, as a near-perfect knowledge transmitter, has rapidly devalued the knowledge-delivery function of traditional classrooms. Teachers find themselves caught between the narrowness of their own specialised training and the explosive, fast-moving breadth of AI, while student engagement continues to decline. In response to this crisis, the author's school officially launched a teaching reform in the spring semester of 2026, shifting its core approach from "knowledge-point instruction" to "project-based learning" (PBL). For the general-education course, which enrols a large number of students from social science and humanities backgrounds, the reform emphasises individual creation using off-the-shelf AI tools, aims at developing a perceptual understanding of AI principles, and involves minimal or no coding. For the computer science majors, in contrast, the advanced course adopts more technically intensive, code-based projects. This paper describes in detail the initial implementation and emerging challenges of this differentiated reform, and reflects on the necessity and pathways for transforming the teacher's role from "knowledge authority" to "learning environment designer."

Refusal geometry reflects refusal training: diverse refusal prefixes can raise stable rank and weaken refusal vector ablation attacks

Refusal training protects AI models from jailbreaks by training models to decline unsafe queries, reducing the risk of misuse. Recent work finds that refusal behavior in aligned language models can be mediated by a single activation direction or a low-dimensional refusal subspace shared across harmful prompts: ablating those directions suppresses refusals while largely preserves other model capabilities. Yet it remains unclear why safety-critical features in a wide range of models emerge in a concentrated, low-dimensional structure. In a case study of OLMo-2-0425-1B-Instruct we find that the refusal geometry reflects refusal training: activation updates resulting from refusal-completion first-token losses explain the resulting refusal direction and refusal subspace. We study refusal directions through the training dynamics across refusal datasets and reveal that their brittleness is associated with repetitive refusal starts, which in turn is linked to concentration of gradients and refusal features in a low-dimensional subspace. Across frozen-model analyses and controlled synthetic fine-tuning, we find evidence of a hardening lever: diverse refusal starts can raise stable ranks of gradients and activation changes, making refusals harder to remove with a vector ablation attack.

Ask Twice, Look Twice: Prompt Echoing Resolves the Question-First Paradox in Vision-Language Models

Where should the question go in a vision-language model (VLM) prompt: before the image or after it? Intuition says before: knowing what is asked should tell the model where to look. Yet across visual question answering benchmarks, question-first prompting consistently underperforms the image-first ordering recommended for frontier VLMs, a phenomenon we term the question-first paradox. We trace this paradox to a conflict between two stages of VLM computation. Logit-lens and attention probes show that question-first prompting steers perception, shifting image patch representations toward question-relevant concepts. But downstream, stranded behind hundreds of image tokens, the question is barely attended by the answer token, which instead commits to image-driven, often wrong answers. Causal attention knockout confirms that the answer reads the question only when it follows the image. This diagnosis yields a training-free fix: question echoing, restating the question on both sides of the image so one copy steers perception while the other is available at answer time. A similar division of labor appears in a fifty-year-old finding on human 'adjunct questions', where repeating a question before and after a passage improves comprehension. Echoing the image as well brings further gains by restoring the whole-image view otherwise lost by a causal decoder. The paradox holds across five open VLMs, costing up to 17.5 group-accuracy points. Echoed prompts recover most of the gap and, on NaturalBench and Winoground, surpass the best single-pass ordering by up to 19 group-accuracy points on Winoground, with no training, fine-tuning, or architecture change. The paradox reveals a tension between steering what a model sees and preserving access to what it was asked; echoing resolves this through prompt design. Project Page: https://rakshanda-cmu.github.io/ask-twice-look-twice/

Learning in Curved Weight Space:Exponential-Linear Weight Reparameterization for Improved Optimization

Many neural networks operations have a multiplicative nature rather than additive: halving or doubling a norm are analogous relatively but require unequal optimization distances when taking linear steps. Adaptive optimizers such as Adam normalize updates per coordinate, but update steps remain additive; weights with very different magnitudes receive similarly sized absolute changes, producing very different relative perturbations. We introduce \textbf{\method} (\textbf{\methodshort}), a weight reparameterization for neural networks that combines a sign-aware symmetric-exponential pathway with an identity-like linear pathway. The symmetric-exponential pathway is near-linear for small raw weights but increasingly curved at larger magnitudes. Additive updates in logarithmic space map to magnitude-proportional changes in effective weight space. The linear pathway provides a direct route through the transform that we hypothesize stabilizes optimization, while learnable scale, curvature, and offset parameters control balance between pathways and the curvature of the exponential pathway. These components create a curved parameter-space geometry that empirically improves speed of loss descent over standard linear parameterization. We also identify a useful \emph{mismatched initialization}: raw weights are chosen so a symmetric version of the transform matches Xavier statistics, but training uses an asymmetric forward transform that leaves positive weights at full strength while making negative weights smaller in magnitude; in small-model ablations, this improves early optimization and may act as a form of symmetry breaking. We train transformers on OpenWebText over nine width$\times$depth configurations, \methodshort reaches matched validation loss in 1.32--1.49$\times$ fewer training steps, with the largest widths seeing the biggest gains.

LLM-Based Test Oracles: Source-of-Authority Taxonomy -- A Systematic Literature Review

Large language models (LLMs) increasingly decide whether software behaves correctly, either by writing a test oracle or by acting as one. Yet two oracles can look identical and rest on different ground: one assertion encodes a written specification, another only what the model learned in training. Prior secondary studies sort oracles by form or by technique, rarely by the property that governs how far a verdict can be trusted: where its authority comes from. This systematic literature review, reported under the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 guidelines, screens 2,436 records to 54 included studies, extended by citation searching (snowballing) to 83 in total. We read the corpus along three axes: the source of an oracle's authority, the form it takes, and the mechanism that adjudicates it. Just over half of the corpus reaches a verdict with no specification at all. That is what lets these oracles work on code with no specification to consult, and what leaves a challenged verdict with less to fall back on. Source and mechanism cross-cut rather than coincide, so a label such as LLM-as-a-judge names how a verdict is produced, not why it should be trusted. Oracle quality is most often judged by resemblance to a known oracle rather than by whether injected faults are caught. The first question to ask of any LLM oracle is therefore what one would point to in defending its verdict. The protocol, search query, and per-study coding sheet are released.

KARMA: Knowledge graph-based Automated Reasoning Materialization and Alignment

Template-based contrastive synthesis is scalable, but its candidates often differ only in a few entity-slots while sequence-level optimization spreads supervision over mostly shared templates. We formalize this as the Resolution Mismatch Problem and propose KARMA, which enumerates schema-constrained paths over domain knowledge graphs and verbalizes them into slot-aligned contrastive candidates. Slot-Parallel Alignment (SPA) then applies a decoupled slot-level objective to route preference supervision to discriminative entity-slots, with slot-aware masked attention serving as an optional packed-evaluation implementation. Across biomedical, computer-science, and chemistry benchmarks, KARMA outperforms base LLM and same-data SFT baselines, and compares favorably with sequence- and token-level preference methods.

SNAP-FM: Sparse Nonlinear Accelerated Projection for Physics-Constrained Generative Modeling

Generative models have emerged as scalable surrogates for physical simulation, yet they offer no guarantee that their outputs respect the conservation laws, boundary conditions, and nonlinear invariants that govern the underlying physics. Constrained sampling closes this gap, enforcing such constraints exactly at inference time without retraining, but at a computational cost: projection, correction and trajectory-optimization steps are repeated during sampling, with these steps becoming expensive for nonlinear constraints. Standard ML frameworks exacerbate this: their dense tensor algebra and limited sparse solver composability obscure the structure that physical constraints naturally induce, making efficient batched nonlinear optimization difficult to realize in practice. We address this bottleneck by exploiting the structure that sample-wise batching and local PDE couplings induce in the projection subproblems -- namely, block-sparse Jacobian and KKT systems -- exposing this structure using ExaModels.jl and solving the resulting sparse nonlinear programs with MadNLP.jl and GPU sparse factorization. Applied to Physics-Constrained Flow Matching (PCFM), on PDE benchmarks with linear, nonlinear, one-dimensional, and two-dimensional constraints, this approach accelerates nonlinear constraint projection while maintaining constraint satisfaction. These results show that sparse GPU nonlinear optimization is a practical foundation for constrained generative sampling in scientific machine learning.