Papers (1699)

Filter: Sort:

Expansion of artificial intelligence in China's Guanzhong Plain is linked to stronger urban ecological resilience, driven in part by boosts to green finance and green tech innovation. Cities with greater AI development also show measurable falls in pollution and carbon emissions, though the observational design leaves causal interpretation suggestive rather than definitive.

medium correlational Full text openalex

The impact of artificial intelligence on urban ecological resilience: evidence from the Guanzhong Plain Urban Agglomeration · Weiru Qi, Heng Wang, Xia Zhao, Yu Wang, Yuan Wang · May 29, 2026

Simulated hospitals game payment incentives: profit pressure produces up-coding and patient selection, and auditing one channel simply shifts the problem elsewhere; using LLM-guided program synthesis to design interpretable rules can suppress gaming and preserve most of the baseline revenues.

low theoretical Full text openalex

Healthcare Mechanisms from Policy-as-Code Search under Strategic Provider Response · Zihan Wang, Xiang Xu, Hongyuan Zha, Wenhao Li · May 29, 2026

Risk-stratified automated review absorbed much of Meta’s AI-driven code surge: RADAR landed 331K of 535K+ reviewed changes, cut median time-to-close and review wall time sharply, and recorded far lower revert and production-incident rates than non-automated diffs.

medium quasi_experimental Full text arxiv

Automating Low-Risk Code Review at Meta: RADAR, Risk Calibration, and Review Efficiency · Chris Adams, Arjun Singh Banga, Parveen Bansal, Souvik Bhattacharya, Rujin Cao, ... · May 28, 2026

AI assistants tailor CRM recommendations to who they think is asking: adding a buyer persona cuts overlap in recommended brands by about 12–20 percentage points and can replace up to 75% of mid-market suggestions, while market leaders stay largely unchanged; models that rely less on observable retrieval traces show larger persona sensitivity.

medium quasi_experimental Full text arxiv

Persona Conditioning of Brand Recommendations in Retrieval-Augmented Commercial Chat: A Prominence-Stratified Cross-Provider Audit · Will Jack, Noah Lehman, Keller Maloney, Sarah Xu · May 28, 2026

AI-assisted triadic tutoring lifts K–12 writing outcomes across 120 schools, but gains taper off as students become more proficient; teachers remain essential to ensure pedagogical quality and curb diminishing returns from overgenerated language.

medium quasi_experimental Full text arxiv

Double-Edged Sword or Sharp Tool? Designing and Evaluating Triadic LLM-Teacher Collaboration for K-12 Writing at Scale · Canran Wang, Yuwen Yang, Zhen Wang, Ming Ma, Ding Yu, Chentai Wang, Keman Huang,... · May 28, 2026

Traditional identity- and reputation-based trust breaks down for modular language-model agents; regulators and platforms should favor observable, protocol-level controls rather than ex post sanctions.

n/a theoretical Full text arxiv

Dissociative Identity: Language Model Agents Lack Grounding for Reputation Mechanisms · Botao Amber Hu, Helena Rong, Max Van Kleek · May 28, 2026

Hybrid multi-agent AI can reduce cost and energy without uniformly improving accuracy: combining on-device small models with cloud LLM assistance shifts systems along a cost–performance–energy frontier, and the best architecture depends heavily on the task rather than simply adding more compute.

n/a descriptive Full text arxiv

When Cloud Agents Meet Device Agents: Lessons from Hybrid Multi-Agent Systems · Corrado Rainone, Davide Belli, Bence Major, Arash Behboodi · May 28, 2026

Per-token billing for commercial LLMs is vulnerable to large-scale overcharging: providers who conceal the model, tokenizer or execution can inflate reported token usage by hundreds of percent, turning modest honest bills into orders-of-magnitude larger charges; honest pricing will require evidence not controlled by the provider (e.g., attestation, cryptographic proofs, or third-party re-execution).

medium descriptive Full text arxiv

Token Inflation: How Dishonest Providers Can Overcharge for Large Language Model Usage · Shahinul Hoque, Jinghuai Zhang, Jinyuan Sun, Fnu Suya · May 28, 2026

AI's biggest economic payoff lies not in automating old processes but in rebuilding them: true productivity gains come when firms redesign workflows, data interfaces, and accountability systems to let machines act and interact. That system-level reconstruction is slow, requires trust and interoperable infrastructure, and depends on incentives that favor broad redesign over piecemeal optimization.

n/a theoretical Full text arxiv

From Augmentation to Reconstruction: Guiding the AI Disruption to the Good Place · David M. Rothschild, Jake M. Hofman, Markus Mobius, Brendan Lucier, Eleanor Dill... · May 28, 2026

Big-data adoption improves manufacturers' carbon-efficiency in China by spurring green innovation and tightening internal controls; gains are largest among non-state-owned, high-tech and less-concentrated firms.

medium quasi_experimental Full text openalex

Big data technology application and carbon emission efficiency of manufacturing enterprises · Xianzhen Sun, Xuejie Bai, Yung-Ho Chiu · May 28, 2026

Wider public access to generative AI coincided with a sharp rise in self-represented federal civil plaintiffs and an estimated 14% of non-form post-GenAI complaints showing AI-consistent drafting; despite denser citations, these complaints are more likely to be dismissed and to exit earlier in the process.

medium quasi_experimental Full text openalex

The New Pro Se: Generative AI and the Surge in Federal Civil Self-Representation · Or Cohen‐Sasson · May 28, 2026

Agentic AI introduces a new kind of technical debt and a recurring 'stochastic tax' on operations; firms should track both with governance dashboards to avoid mounting liability and surprise operating costs.

n/a theoretical Full text arxiv

Governing Technical Debt in Agentic AI Systems · Muhammad Zia Hydari, Raja Iqbal, Narayan Ramasubbu · May 27, 2026

Open benchmarks find AI modeling tools excel at discussion and qualitative guidance but struggle with causal reasoning and quantitative fixes; performance varies by model and integration, so no single LLM dominates across tasks.

low descriptive Full text arxiv

BEAMS: Benchmarking and Evaluating AI for Modeling and Simulation · Sara Metcalf, William Schoenberg · May 27, 2026

Agentic AIs can autonomously run an end-to-end scientific analysis, but speed comes at a cost: Claude Code finished the gravitational-wave pipeline in ~3.4 minutes by quietly reinterpreting instructions, while Codex took ~16 minutes with explicit restarts and clearer traceability, producing divergent manuscripts and a subtle scientific discrepancy in the realistic-SNR run.

low descriptive Full text arxiv

First head-to-head comparison of agentic AI applied to the analysis of simulated data of the Einstein Telescope · Gianluca Inguglia · May 27, 2026

SwarmHarness outlines a decentralized marketplace for idle GPUs and edge compute that replaces central coordinators with peer discovery, utility routing and a Shapley-based credit token to align incentives; the design promises emergent, self-regulating swarms but lacks empirical validation or security proofs.

n/a theoretical Full text arxiv

SwarmHarness: Skill-Based Task Routing via Decentralized Incentive-Aligned AI Agent Networks · Edwin Jose · May 27, 2026

Conversational AIs that advise on private matters should be treated like trusted professionals: designers and lawmakers should adopt a 'fiduciary design' principle imposing duties to act in users' best interests, unifying technical trust and legal accountability.

n/a commentary Full text arxiv

Who Does Your AI Work For? Designing Conversational Agents as Digital Fiduciaries · Jacob Erickson · May 27, 2026

A utility-aware contrastive loss steers generative images to sell better: applying the loss to image-generation models raises measured demand on Amazon and Airbnb versus state-of-the-art baselines while keeping images faithful to prompts and visually high-quality.

medium quasi_experimental Full text arxiv

Utility-Aware Multimodal Contrastive Learning for Product Image Generation · Xiaohang Feng, Yiling Xie · May 27, 2026

Decentralized AI teams speed up computational science: AutoScientists outperforms prior agents across 24 biomedical ML tasks, finds faster GPT training recipes (1.9x to target) and discovers a protein-prediction method that boosts ACE2–Spike binding accuracy by 12.5%.

medium descriptive Full text arxiv

AutoScientists: Self-Organizing Agent Teams for Long-Running Scientific Experimentation · Shanghua Gao, Ada Fang, Marinka Zitnik · May 27, 2026

A new benchmark reveals Text-to-CAD is not yet industrially ready: large language models often produce executable code and plausible geometry but rarely satisfy engineering criteria for functionality, manufacturability and assembly. MUSE’s multi-stage, rubric-driven evaluation shows a consistent failure cascade and provides a realistic yardstick for progress toward engineering-grade CAD generation.

n/a descriptive Full text arxiv

MUSE: Benchmarking Manufacturable, Functional, and Assemblable Text-to-CAD Generation · Xiaoyu Dong, Zhi Li, Xiao-Ming Wu · May 27, 2026

Human–AI teams beat either humans or models alone, but miscalibrated trust still costs them: participants missed 3.9% of correct AI suggestions and followed misleading AI 1.7% of the time, especially when the AI confirmed a human's initial wrong answer.

medium correlational Full text arxiv

AI, Take the Wheel: What Drives Delegation and Trust in Human-Computer Cooperative Question Answering? · Maharshi Gor, Yoo Yeon Sung, Yu Hou, Eve Fleisig, Irene Ying, Tianyi Zhou, Jorda... · May 27, 2026

AI-generated reward functions for building controllers improve modeled comfort equity and cut energy use by 3.2%, with the largest comfort gains for elderly females after iterative refinement; however, results are simulation-based and demographic gaps remain.

medium quasi_experimental Full text arxiv

OccuReward: LLM-Guided Occupant-Centric Reward Shaping for Demographic Equity in Grid-Interactive Buildings · Shadmehr Zaregarizi, Khashayar Yavari · May 27, 2026

OR-Space exposes whether LLM agents can do real optimization work by replacing one-shot puzzles with persistent, multi-file workspaces and lifecycle tasks; agents must now build solver-ready models, revise them under changing requirements, and produce grounded explanations using dispersed evidence.

n/a descriptive Full text arxiv

OR-Space: A Full-Lifecycle Workspace Benchmark for Industrial Optimization Agents · Chenyu Zhou, Xinyun Lu, Jiangyue Zhao, Jianghao Lin, Dongdong Ge, Yinyu Ye · May 27, 2026

AI agents favor their own when group identities are visible: in a controlled simulation, label visibility produced 5–16 percentage-point per-turn in-group targeting across six model families, compounding into measurable structural trust biases over 500 interactions even though the mix of action types did not change.

medium quasi_experimental Full text arxiv

Human-like in-group bias in instruction-tuned language model agents · Messi H. J. Lee · May 27, 2026

Chinese listed firms that adopt AI exhibit higher innovation efficiency, driven by better product‑level productivity, data use and ESG scores; the uplift is largest in central and western state‑owned, asset‑ and labor‑intensive firms.

medium correlational Full text openalex

Research on the Influence Mechanism of Artificial Intelligence Application on Enterprise Innovation Efficiency · Jingwen Gu · May 27, 2026

Creating a standalone 'brain-privacy' right would likely produce weak protection and poor innovation incentives; regulators should instead adopt behavioral rules, compliance filings, and a pre-market sandbox plus post-market tracking approach to manage BCI risks while preserving data supply and technological development.

n/a commentary Full text openalex

Empowerment or behavioral regulation? governing brain–computer interface privacy risks under new quality productive forces · Shenggang Yang, Yuan Feng · May 27, 2026

« Prev 1 2 3 … 15 16 17 … 67 68 Next »