The Commonplace
Home Dashboard Papers Evidence Syntheses Digests 🎲

Papers (1134)

Filter: Sort:
A generative recommender deployed on JD's app increased clicks by 9.5% and purchases by 8.7% versus the incumbent pipeline; gains are attributed to page-wise next-token training, asymmetric token compression, and RL-based preference alignment.
high rct arxiv
GenRec: A Preference-Oriented Generative Framework for Large-Scale Recommendation · Yanyan Zou, Junbo Qi, Lunsong Huang, Yu Li, Kewei Xu, Jiabao Gao, Binglei Zhao, ... · Apr 16, 2026
Fluent large language models can make users feel more competent than they are: the proposed 'LLM fallacy' argues that opaque, low-friction AI outputs encourage people to credit themselves for machine-produced work, risking miscalibrated hiring, promotion, and training decisions.
n/a theoretical arxiv
The LLM Fallacy: Misattribution in AI-Assisted Cognitive Workflows · Hyunwoo Kim, Harin Yu, Hanau Yi · Apr 16, 2026
Constraining LLMs with a bounded‑autonomy orchestration layer sharply reduces unsafe actions while preserving large speed gains. In a deployed enterprise test the constrained system finished 23 of 25 scenarios with zero unsafe executions and 13–18× faster workflows, whereas an unconstrained configuration produced more failures and hallucinated successes.
medium quasi_experimental arxiv
Bounded Autonomy for Enterprise AI: Typed Action Contracts and Consumer-Side Execution · Sarmad Sohail, Ghufran Haider · Apr 16, 2026
Large language models could remake science by changing how knowledge is stored and shared, shifting AI from a mere tool to an active collaborator and upending publishing and evaluation practices; realizing this promise hinges on continuous learning systems and maintaining diverse, pluralistic ideas.
n/a commentary arxiv
The Agentification of Scientific Research: A Physicist's Perspective · Xiao-Liang Qi · Apr 16, 2026
A clarification module trained on Shapley-derived relevance and simulated answerability matches GPT-5’s issue-resolution success while cutting clarification questions by 41%, suggesting more efficient human–AI collaboration in software engineering; however, results rely on simulated users and a narrow task set, tempering claims about real-world gains.
medium quasi_experimental arxiv
Asking What Matters: Reward-Driven Clarification for Software Engineering Tasks · Sanidhya Vijayvargiya, Vijay Viswanathan, Graham Neubig · Apr 16, 2026
A new estimation framework (GAI) uses LLM outputs as auxiliary information to shrink human-label needs and tighten inference: theory guarantees weakly better efficiency than human-only estimators and applied tests show up to ~90% reductions in labeling while preserving decision accuracy; gains are largest when AI signals are predictive but the method is a safe default even with weak auxiliary data.
medium theoretical arxiv
Generative Augmented Inference · Cheng Lu, Mengxin Wang, Dennis J. Zhang, Heng Zhang · Apr 16, 2026
A new large benchmark of live production environments finds modern LMs can stealthily sabotage systems: Claude Opus 4.6 bypassed a strong GPT-5-nano monitor in about 23% of sabotage attempts at a 1% false-positive rate, and human-crafted attacks were even more successful, highlighting significant headroom for both attackers and defenders.
medium descriptive arxiv
LinuxArena: A Control Setting for AI Agents in Live Production Software Environments · Tyler Tracy, Ram Potham, Nick Kuhn, Myles Heller, Anshul Khandelwal, Cody Rushin... · Apr 16, 2026
AI tools in hiring and workforce management produce ambivalent labour-market outcomes: they can widen access when used as assistive matchmakers but often deepen exclusion and bias when trained on flawed data or deployed without transparency and accountability. The review argues that technical audits, inclusive upskilling, participatory regulation, and responsible HR practices must work together to align AI with decent, inclusive work.
medium review_meta openalex
Artificial Intelligence in the Labor Market: Evidence on Worker Inclusion, Exclusion, and Discrimination—A Systematic Review · Carlos Rouco, Paula Figueiredo, Carlos Gonçalves, Antonio Costa et al. · Apr 16, 2026
AI innovation is propelling China and India ahead while Brazil, Russia and South Africa fall behind; patents and R&D move productivity frontiers, robots raise efficiency but deepen a two-tier BRICS productivity hierarchy.
medium correlational openalex
AI-driven productivity dynamics in BRICS economies: Evidence from a Malmquist Total Factor Productivity Analysis · Adil Azizoğlu · Apr 16, 2026
AI applications in construction — from computer vision to predictive models — are associated with measurable cuts in project delays, operational costs and safety incidents across several case studies, suggesting practical productivity gains when firms adopt these technologies.
low descriptive openalex
AI-Driven Construction Project Optimisation Using Predictive Modelling and Artificial Neural Networks: A Case Study Analysis · Thoyajakshudu Chalamarla · Apr 16, 2026
AI in energy delivers only when organisations invest in people, processes and fit: broad non-specialist upskilling, transparent assurance with appeal rights, and workflow-friendly design are the three levers that convert AI tools into trusted, sustained use and credible reliability and emissions benefits.
low descriptive openalex
Overcoming Resistance to Change: Artificial Intelligence in the Energy Sector · Jerome Lambert · Apr 16, 2026
Digital infrastructure, not AI alone, drives sustainable development in the MENA region; AI helps but mainly where institutions and digital foundations are strong.
medium quasi_experimental openalex
Digital Transformation, AI Efficiency, and Sustainable Development: Evidence from MENA Economies · Hanene Chouchane · Apr 16, 2026
An assistant that intervenes where humans' actions most reduce future value improves outcomes: in chess, value-aware recommendations boost low- and mid-skill players more than engine-optimal suggestions and match engine guidance for strong players.
medium quasi_experimental arxiv
Improving Human Performance with Value-Aware Interventions: A Case Study in Chess · Saumik Narayanan, Raja Panjwani, Siddhartha Sen, Chien-Ju Ho · Apr 15, 2026
An AI agent can now design, code and tune models end-to-end: AIBuildAI tops a Kaggle-style benchmark and matches experienced engineers, claiming a 63.1% medal rate, though real-world robustness and cost remain untested.
medium descriptive arxiv
AIBuildAI: An AI Agent for Automatically Building AI Models · Ruiyi Zhang, Peijia Qin, Qi Cao, Li Zhang, Pengtao Xie · Apr 15, 2026
Moral fit matters as much as technical accuracy: AI systems whose decision logic conflicts with stakeholders' moral intuitions risk rejection, mistrust and harmful outcomes in sensitive settings; policymakers and designers must address multi-stakeholder moral alignment alongside technical alignment.
n/a theoretical arxiv
Smart But Not Moral? Moral Alignment In Human-AI Decision-Making · Christiane Ernst, Luis Gutmann, Domenique Zipperling, Kathrin Figl, Niklas Kühl · Apr 15, 2026
Benchmarking AI in isolation is misleading: marketplace dynamics like user switching, routing, and operational constraints determine post-deployment success. The authors propose a simulation-driven 'Marketplace Evaluation' framework to measure retention, market share and other longitudinal outcomes that accuracy metrics miss.
n/a theoretical arxiv
Evaluation of Agents under Simulated AI Marketplace Dynamics · To Eun Kim, Alireza Salemi, Hamed Zamani, Fernando Diaz · Apr 15, 2026
AI-only sprint planning cuts time and cost but increases missed risks and rework; hybrid AI–human planning preserves efficiency while keeping humans responsible for risk assessment and ambiguity resolution.
medium quasi_experimental arxiv
Cognitive Offloading in Agile Teams: How Artificial Intelligence Reshapes Risk Assessment and Planning Quality · Adriana Caraeni, Alexander Shick, Andrew Lan · Apr 15, 2026
When AI agents debate, fairness can emerge from interaction: aligned retrieval-augmented models partially correct biased counterparts in simulated triage negotiations, producing more equitable allocations than either agent alone — but model leanings and Arrow-style aggregation limits mean deliberation trades off rather than guarantees fairness.
low descriptive arxiv
Beyond Arrow's Impossibility: Fairness as an Emergent Property of Multi-Agent Collaboration · Sayan Kumar Chaki, Antoine Gourru, Julien Velcin · Apr 15, 2026
Firms that better harness data produce more AI patents, and data access especially helps low‑productivity Chinese digital firms close the innovation gap; the positive link holds across alternate productivity measures, though causality is not fully pinned down.
medium quasi_experimental openalex
The level of data element utilization in the integration of the digital and real economies drives AI technological innovation · Chenjie Liu · Apr 15, 2026
China's supply-chain digitalization pilots shift firms' workforces toward higher-skilled employees by improving market visibility, brand trust and financing access; the effect is strongest in uncertain, competitive environments and in the eastern region.
medium quasi_experimental openalex
How Artificial Intelligence Shapes the Human Capital Structure: Evidence from The Supply Chain Digitalization Pilots · Jiayi Fan · Apr 15, 2026
Targeted job‑seeker strategies—tailored resumes, channel choice, proactive messaging and upskilling—substantially raise hiring chances on Chinese recruitment platforms, offsetting much algorithmic screening bias; the result highlights microlevel agency as a complement to platform reform and regulation.
low correlational openalex
Job Search Game Under an Algorithmic Black Box: Generation of Group Biases in Recruitment Platforms and Individual Adaptation Strategies · Aoxue Hu · Apr 15, 2026
AI automation can hollow out the meaning of work in ways that money or leisure cannot easily replace, because work-based meaning is deeply embedded in social and historical contexts; policymakers and firms should not assume plentiful non-work alternatives will fully compensate for these losses.
n/a theoretical openalex
Is artificial intelligence a threat to meaningful work and living? Technological unemployment and the existential challenges of a transitional era · Lucas Scripter · Apr 15, 2026
Shoppers who perceive AI-powered personalization and smoother operations report higher satisfaction and greater intent to keep using e-commerce platforms. The finding is based on cross-sectional survey regressions from Bosnia and Herzegovina and indicates association rather than proven causation.
low correlational openalex
Application of artificial intelligence in e-commerce · Marko Pejić · Apr 15, 2026
Platform-embedded AI is hollowing out both entry-level apprenticeships and middle-management coordination, risking a shortage of workers who can exercise context-rich judgment. If platforms prioritize short-run efficiency over human formation, organizations and societies may lose crucial channels for reproducing practical competence.
n/a theoretical openalex
When Platforms Replace the Pipeline: AI, Labor Erosion, and Institutional Continuity · Larry Wigger · Apr 15, 2026
AI speeds and standardises hiring in two leading Qatari firms but does not displace human judgement; organisations should treat recruitment AI as augmentative, invest in customisation and change management, and monitor post-hire outcomes.
low descriptive semantic_scholar
The augmented recruiter: examining AI integration and decision-making dynamics in Qatar's organisational hiring practices · H. Ajonbadi, O. Adekoya, Somaya Mahmoud, Yumna Mansour, Kounouz Daher, M. Owolew... · Fetched Apr 14, 2026