The Commonplace
Home Papers Evidence Explore Trends Syntheses Digests About 🎲 Workforce Futures

Papers (1747)

Filter: Sort:
Lab-style benchmarks and common NLP scores miss systemic failures when AI agents run continuously in production: four of seven production-specific failure modes were invisible to ROUGE, BERTScore and standard agentic benchmarks, and the authors offer PAEF, a continuous, production-oriented evaluation framework and open-source implementation to close the detection gap.
medium descriptive Full text arxiv
Evaluating Agentic AI in the Wild: Failure Modes, Drift Patterns, and a Production Evaluation Framework · Mukund Pandey · May 02, 2026
Cheaply copied, easily deployed AI can aggregate decision-making power into single highly efficient nodes, creating a one-stop source of irreversible system-level harm; safety should therefore prioritize institutional and technical boundary stabilization — layered control, authorization, and externally reviewable limits — not solely perfect alignment.
n/a theoretical Full text arxiv
AI Safety as Control of Irreversibility: A Systems Framework for Decision-Energy and Sovereignty Boundaries · Wesley Shu, Peng Wei · May 02, 2026
AI agents could sever the centuries-old link between population trends and GDP growth by acting as economic actors in their own right, creating an expanding 'algorithmic' population and new pressures on pensions, taxation and international competitiveness. Policymakers must measure agent productivity (cFTE), energy use (AEP) and reconsider fiscal and market rules to manage the geoeconomic and social consequences.
n/a theoretical Full text semantic_scholar
DIGITAL AGENTS AS FUNCTIONAL EQUIVALENTS OF ECONOMIC ACTORS: THE DECOUPLING OF DEMOGRAPHICS AND ECONOMIC GROWTH · N. Reznikova, V. Panchenko, M. Grod · Fetched May 02, 2026
Digital-health technologies—including EHRs, telemedicine, AI and IoT—are improving efficiency in Indian hospitals, but benefits are concentrated in cities; lack of funding, poor infrastructure and regulatory gaps prevent widespread adoption in rural areas.
medium review_meta Full text semantic_scholar
A Comprehensive Review of Technology Adoption and Its Impact on Organisational Productivity in the Healthcare Industry in India · Manali Gupta, Dr. Garima Bharadwaj, Dr. Pooja Tiwari · Fetched May 02, 2026
AI development at Chinese listed firms widens wage gaps: firms with more AI-related patents exhibit higher skill premiums and lower labor income shares, with stronger effects in non-state firms, more digitalized firms, and less concentrated industries.
medium quasi_experimental Summary openalex
The Impact of Artificial Intelligence on the Labor Skill Premium: Evidence from Chinese Listed Companies · Hui Liang, Xuxia Zhang, Jingbo Fan · May 02, 2026
A theoretical model finds the AI computing-power industry moves through distinct stages—early 'natural monopoly', mid-period 'regulatory stalemate' traps, and a mature phase of co-opetition—shaped by downstream vertical integration and incumbents' choice to open interfaces, suggesting time-varying, targeted antitrust interventions are required.
n/a theoretical Summary openalex
Evolutionary Dynamics of Openness, Dependence, and Regulation in AI Computing Power Innovation Ecosystem · Zhengrui Li, Qingjin Wang, Shuai Huang, Lan Tian · May 02, 2026
AI coding assistants often boost measurable developer activity but can undermine software reliability — a 'Productivity‑Reliability Paradox' driven by non-deterministic code generation and poor specification governance. A literature synthesis, taxonomy and a short pilot suggest strengthening specification discipline (not only improving models) is the key lever for dependable AI-augmented software development.
low review_meta Full text arxiv
The Productivity-Reliability Paradox: Specification-Driven Governance for AI-Augmented Software Development · Sabry E. Farrag · May 01, 2026
Treat the controller, not the model, as Bayesian: applying Bayesian decision theory at the orchestration layer can make agentic AI systems better at choices under uncertainty—keeping beliefs calibrated and actions utility-aware—while avoiding the computational burden of fully Bayesian LLMs.
n/a theoretical Full text arxiv
Position: agentic AI orchestration should be Bayes-consistent · Theodore Papamarkou, Pierre Alquier, Matthias Bauer, Wray Buntine, Andrew Daviso... · May 01, 2026
Retrieval-augmented chat assistants boost human task accuracy across small and large LLMs; but users don't rate larger models as more usable or satisfying, revealing a gap between technical performance gains and user perception.
medium rct Full text arxiv
Seeking Information with RAG-Assistants: Does Model Size Matter in Human-AI Collaborations? · Lennard C. Froma, Tom Kouwenhoven, Maaike H. T. de Boer, Catholijn M. Jonker, Ma... · May 01, 2026
Italian firms that adopt AI see higher productivity and profits and shift hiring toward white-collar roles, while overall employment remains unchanged; adoption is currently concentrated among larger, knowledge‑intensive firms.
medium quasi_experimental Summary semantic_scholar
The economic impact of artificial intelligence: evidence from Italian firms · Tiziano Ropele, A. Tagliabracci · Fetched May 01, 2026
AI is restructuring labour markets: routine tasks are being automated while demand rises for complex cognitive and social skills, creating both productivity gains and transitional risks; targeted reskilling and policy frameworks are necessary to ensure inclusive outcomes.
n/a review_meta Summary semantic_scholar
AI and the Transformation of Human Employment: Challenges, Opportunities and Future Prospects · B. Harshini, Dr.K.B. Ravindra, Dr. Suresh Kurumalla · Fetched May 01, 2026
AI-driven automation threatens semi- and unskilled jobs across Sub‑Saharan Africa, with the greatest risks in manufacturing; without targeted upskilling, labour-market reforms and equitable AI preparedness the technology may deepen unemployment and inequality.
low review_meta Summary semantic_scholar
The Impact of AI-Driven Automation on Semi and Unskilled Workers in Sub-Saharan Africa's Manufacturing Sector: A Socio-Economic Analysis and Policy Perspective · Dr Vusi S. Mncube · Fetched May 01, 2026
Creators hide their use of generative AI to preserve authenticity, turning 'efficiency' gains into downstream repair and performance work; those with more education, money, or team support are better able to pass, widening trust-based inequality.
low descriptive Full text semantic_scholar
AI passing and invisible authenticity labor: trust vulnerabilities and inequality in China's short-video creator economy · Xiaolong Su, Tingli Liu, Zhuoqi Ye, Yixue Mei · Fetched May 01, 2026
AI lifted Russia's potential GDP in 2023–2025, driven primarily by productivity gains and process optimisation; however, benefits were uneven across industries and limited by digital readiness, human-resource shortages and institutional constraints.
low descriptive Summary semantic_scholar
THE IMPACT OF AI ON POTENTIAL GDP AND LONG-TERM ECONOMIC GROWTH · Karina Yu. Sherstyukova, V. S. Tormozov, S. Syurkova · Fetched May 01, 2026
A tokenized, self-devaluing commodity currency is proposed as a global 'honest money' for the AI era; the design aims to protect cross-border AI capital allocation from fiat depreciation but remains untested and faces legal, monetary-policy and adoption barriers.
n/a theoretical Full text arxiv
RSDM: The Consensus Honest Money in the AI Era · Boliang Lin, Ruixi Lin · May 01, 2026
Smaller open-weight models can perform the bulk of routine agent tasks faster and cheaper than frontier models, but complex long-horizon planning still favors the largest systems; a hybrid design — small models for routine actions, big models for deep planning — is a practical way to build agentic pipelines.
medium descriptive Full text arxiv
AgentFloor: How Far Up the tool use Ladder Can Small Open-Weight Models Go? · Ranit Karmakar, Jayita Chatterjee · May 01, 2026
Endpoint-level differences — not just model families — drive large swings in accuracy, latency and cost: the same model served from different endpoints can vary by over 10 percentage points in accuracy, an order of magnitude in tail latency, and multiple-fold in energy and dollars per correct answer, changing which endpoints look best once realistic workloads are considered.
medium descriptive Full text arxiv
Token Arena: A Continuous Benchmark Unifying Energy and Cognition in AI Inference · Yuxuan Gao, Megan Wang, Yi Ling Yu · May 01, 2026
AI's productivity dividend is conditional: in Chinese listed firms, AI adoption substantially raises TFP for resource‑constrained firms but yields negligible or diminishing returns for technologically advanced firms in hypercompetitive markets, driven by automation efficiencies in the former and redundant AI/innovation stagnation in the latter.
medium correlational Full text openalex
The Heterogeneous Effects of Artificial Intelligence on Enterprise Total Factor Productivity: Key Mechanisms and Strategic Implications · Xu Chu, Qingyu Han · May 01, 2026
Banks that integrate AI see measurable gains in profitability and governance efficiency — higher ROA, improved operational and risk-adjusted performance, and lower compliance costs — but algorithmic bias and employee displacement remain material risks.
medium quasi_experimental Full text openalex
Research on the Transformation Acceleration of Financial Institutions and Governance Efficiency with Artificial Intelligence Technology · Guangyuan Han, Huiping Shang, Keying Li, Jingzhi Han · May 01, 2026
Synthetic 'computers' simulate month-long, user-specific work environments to train agents: 1,000 long-horizon runs (8+ hours, ~2,000 turns) produce richer experiential signals and improve agent performance on in- and out-of-domain productivity evaluations, though real-world validation is still pending.
low descriptive Full text arxiv
Synthetic Computers at Scale for Long-Horizon Productivity Simulation · Tao Ge, Baolin Peng, Hao Cheng, Jianfeng Gao · Apr 30, 2026
A live benchmark shows state-of-the-art LLM agents complete at most two-thirds of realistic workflow tasks: the top model passes 66.7% of 105 controlled tasks. Failures cluster in HR, management and multi-system business workflows, indicating end-to-end workflow automation remains far from solved.
n/a descriptive Full text arxiv
Claw-Eval-Live: A Live Agent Benchmark for Evolving Real-World Workflows · Chenxin Li, Zhengyang Tang, Huangxin Lin, Yunlong Lin, Shijue Huang, Shengyuan L... · Apr 30, 2026
Clinician overrides are valuable preference signals, not noise; jointly learning a reward model and clinician capability from longitudinal override and outcome data can align clinical AI to patient trajectories and prevent the systematic suppression of correct-but-difficult recommendations. Value-based chronic care—with dense longitudinal outcomes and natural capability variation—offers especially favorable conditions for this approach.
n/a theoretical Full text arxiv
Learning from Disagreement: Clinician Overrides as Implicit Preference Signals for Clinical AI in Value-Based Care · Prabhjot Singh, Abhishek Gupta, Chris Betz, Abe Flansburg, Brett Ives, Sudeep La... · Apr 30, 2026
Language models, when paired with symbolic abstractions and numerical optimisation, substantially improve simulated linkage design—cutting geometric error by up to 68% and markedly raising structural validity; however, evidence is limited to simulation and a small set of targets.
medium descriptive Full text arxiv
Language Models Refine Mechanical Linkage Designs Through Symbolic Reflection and Modular Optimisation · João Pedro Gandarela, Thiago Rios, Stefan Menzel, André Freitas · Apr 30, 2026
Pictures and colours nudge vision-language models: exposure to ‘kind’ images and cooperative color cues increases model cooperation in an iterated Prisoner’s Dilemma, and prompt edits, Chain-of-Thought, or reducing visual tokens can lessen—but not uniformly eliminate—these biases across models.
medium quasi_experimental Full text arxiv
The Effects of Visual Priming on Cooperative Behavior in Vision-Language Models · Kenneth J. K. Ong · Apr 30, 2026
Differentiable AI weather models can produce gradient-based payment signals that recover near-optimal sensor value and support monotonic rewards for participatory IoT networks; however, these signals are computationally vulnerable to adversarial input inflation and need external baselines for reliable detection.
medium descriptive Full text arxiv
Calibrating Attribution Proxies for Reward Allocation in Participatory Weather Sensing · Mark C. Ballandies, Michael T. C. Chiu, Claudio J. Tessone · Apr 30, 2026