The Commonplace
Home Dashboard Papers Evidence Syntheses Digests 🎲

Papers (1134)

Filter: Sort:
Refinement loops often hurt unless the model almost never introduces new errors: iterative self-correction degrades accuracy for many LLMs unless EIR is near zero, but a lightweight 'verify-first' prompt can reduce EIR and restore gains on affected models.
medium quasi_experimental arxiv
When Does LLM Self-Correction Help? A Control-Theoretic Markov Diagnostic and Verify-First Intervention · Aofan Liu, Jingxiang Meng · Apr 24, 2026
Revealing a few tailored features to humans — rather than giving full predictions — can boost human–algorithm teamwork, but designing highlights for fully rational users is computationally infeasible; optimizing for more realistic, 'naive' users yields tractable and robust gains in calibrated housing-data simulations.
n/a theoretical arxiv
Algorithmic Feature Highlighting for Human-AI Decision-Making · Yifan Guo, Jann Spiess · Apr 24, 2026
A learning-augmented robot ran a live motor assembly station for over five hours, producing 108 motors with a 99.4% quality pass rate and near-human cycle times while operating without physical fencing.
medium descriptive arxiv
Learning-augmented robotic automation for real-world manufacturing · Yunho Kim, Quan Nguyen, Taewhan Kim, Youngjin Heo, Joonho Lee · Apr 24, 2026
Machine-learning contests invite strategic 'benchmark hacking': lower-skill entrants divert effort into task-specific tuning rather than true capability gains, while higher-skill entrants do not. Concentrating prizes among top finishers reduces such gaming and yields better generalization.
medium theoretical arxiv
On Benchmark Hacking in ML Contests: Modeling, Insights and Design · Xiaoyun Qiu, Yang Yu, Haifeng Xu · Apr 24, 2026
Human–AI relations should be designed as governed co-evolution, not obedience: a formal 'conditional mutualism' model shows reciprocal, regulated coupling produces stable coexistence and fairer outcomes, whereas ungoverned interactions risk lock-in, fragility and domination.
n/a theoretical arxiv
A Co-Evolutionary Theory of Human-AI Coexistence: Mutualism, Governance, and Dynamics in Complex Societies · Somyajit Chakraborty · Apr 24, 2026
Bias in AI hiring often emerges from interactions across fragmented vendor–platform–employer supply chains, not single components; regulators and firms should mandate cross-supplier documentation, system-level audits, and continuous monitoring to restore meaningful accountability.
n/a review_meta openalex
How Supply Chain Dependencies Complicate Bias Measurement and Accountability Attribution in AI Hiring Applications · Gauri Sharma, Maryam Molamohammadi · Apr 24, 2026
Fix a numerical safety target, then test: the paper proposes RoMA/gRoMA black‑box statistical tests that let developers produce auditable upper bounds on AI failure rates without revealing model internals, enabling conformity assessments aligned with laws like the EU AI Act.
n/a theoretical arxiv
Bounding the Black Box: A Statistical Certification Framework for AI Risk Regulation · Natan Levy, Gadi Perl · Apr 23, 2026
AEL — a two-timescale approach pairing bandit-guided memory retrieval with LLM-driven reflection — lifts trading-agent Sharpe to 2.13 and reduces variance on a 208-episode portfolio benchmark, beating prior self-improving methods; notably, adding further architectural complexity harms performance, suggesting the bottleneck is diagnosing how to use past experience rather than more components.
medium other arxiv
AEL: Agent Evolving Learning for Open-Ended Environments · Wujiang Xu, Jiaojiao Han, Minghao Guo, Kai Mei, Xi Zhu, Han Zhang, Dimitris N. M... · Apr 23, 2026
Unobserved counterfactuals in sequential decision systems systematically under-expose marginalized groups and can amplify exclusion; modelling model- and feedback-uncertainty and using uncertainty-aware exploration reduces outcome variance for disadvantaged groups without sacrificing expected institutional utility.
low theoretical arxiv
Fairness under uncertainty in sequential decisions · Michelle Seng Ah Lee, Kirtan Padh, David Watson, Niki Kilbertus, Jatinder Singh · Apr 23, 2026
Agentic AI promises sharper market efficiency and liquidity but brings new systemic and regulatory risks; unless governance, monitoring and interpretability tools keep pace, autonomous trading systems could amplify instability.
n/a review_meta arxiv
Agentic Artificial Intelligence in Finance: A Comprehensive Survey · Irene Aldridge, Jolie An, Riley Burke, Michael Cao, Chia-Yi Chien, Kexin Deng, R... · Apr 23, 2026
Coupling LSTM demand forecasts with MILP optimization meaningfully improves supply‑chain performance: forecasting errors fall ~12–15%, inventory cost drops 5.4%, stockouts decline 27.5% and service levels rise to 97.8% in textile and PPE datasets.
medium descriptive arxiv
Hybrid Deep Learning Approach for Coupled Demand Forecasting and Supply Chain Optimization · Nusrat Yasmin Nadia, Md Habibul Arif, Habibor Rahman Rabby, Md Iftekhar Monzur T... · Apr 23, 2026
A Probabilistic Innovation Methodology proposes that organised, real-time crowdsourced R&D can convert open-ended uncertainty into governable probabilistic search, shifting innovation from opaque guessing toward testable, steerable discovery; as AI reduces ideation costs, the paper argues, the primary bottlenecks become coordination, adjudication and validation.
n/a theoretical semantic_scholar
Probabilistic Innovation Methodology: A Scientific Methodology for Real-Time Problem Solving in Complex Systems. Formalising Crowdsourced R&D, Hackathon-Type Search, and Epistemic Mode Transitions · Christian Callaghan · Fetched Apr 23, 2026
Agentic AI could speed chip and physical-design R&D by automating workflow, heuristic discovery and tool use; current evidence is promising but largely limited to prototypes, benchmark fragments and lab demonstrations.
n/a review_meta semantic_scholar
Invited: Agentic AI for Physical Design R&D: Status and Prospects · Amur Ghose, Andrew B. Kahng, Sayak Kundu, Bodhisatta Pramanik · Fetched Apr 23, 2026
Large language models tilt toward government-intervention explanations: across 1,056 ideology-contested economic cases, 18 of 20 models are more accurate when the true effect aligns with intervention-friendly expectations and, when wrong, disproportionately predict intervention-oriented effects, a bias that one-shot prompting does not remove.
medium descriptive arxiv
Ideological Bias in LLMs' Economic Causal Reasoning · Donggyu Lee, Hyeok Yun, Jungwon Kim, Junsik Min, Sungwon Park, Sangyoon Park, Ji... · Apr 23, 2026
As machines become autonomous, market efficiency must be redefined: the authors show that competitive markets can remain Pareto optimal only after welfare is conditioned on agents' autonomy and appropriate delegation and verification institutions are in place; the standard theorem appears as the special case when AI autonomy is negligible.
n/a theoretical arxiv
Post-AGI Economies: Autonomy and the First Fundamental Theorem of Welfare Economics · Elija Perrier · Apr 23, 2026
Digital tools — from IoT to AI — materially strengthen supply-chain resilience by improving visibility, which accounts for roughly two-thirds of the resilience gains; the benefits are largest in complex, technology-intensive industries.
medium quasi_experimental openalex
The Role of Digital Technologies in Enhancing Supply Chain Visibility and Resilience · Selorm Aniwa · Apr 23, 2026
China’s AI pilot zones lift manufacturers’ ESG performance by spurring R&D and tighter environmental compliance; gains are strongest among private and high‑tech firms and where firms allocate resources and operate efficiently.
medium quasi_experimental openalex
The Impact of National New-Generation Artificial Intelligence Innovation and Development Pilot Zone Construction on ESG Performance of Manufacturing Enterprises · Yi Cao, Zhou Lan, Jie Dong, Ling Cao · Apr 23, 2026
AI's impact on jobs is uneven: routine-heavy industries face the greatest automation risk, whereas knowledge sectors see augmentation and rising skill demands. Policymakers and firms must prioritize targeted re-skilling and sectoral risk assessments to manage workforce transitions.
medium review_meta openalex
AI and the Future of Job Profiles: A systematic Review of Sectoral Job Transformation, Risks and Future Impacts · Shrivastava Anshul, P Rohit Kumar, Sharma Anil Kumar · Apr 23, 2026
Intelligent management systems are recasting firms as 'algorithmic enterprises' that can optimize strategy and operations in real time, but they concentrate data power and blur lines of accountability.
n/a theoretical openalex
Algorithmic Enterprises: Rethinking Firm Strategy in the Age of Intelligent Management Systems · Swaraj Dash · Apr 23, 2026
Gender-diverse boards curb corporate tax avoidance, and firms that deploy AI see a larger governance effect; AI capability amplifies the tax‑compliance role of diverse boards in developing‑country firms.
medium quasi_experimental openalex
AI-Enabled Governance: Board Gender Diversity and Corporate Tax Avoidance · Marwan Mansour, Mo’taz Al Zobi, Ahmad Marei, Luay Daoud, Nour Ibrahim Kurdi · Apr 23, 2026
Microsoft 365 Copilot is seen as user-friendly and reliable and yields clear efficiency gains for administrative work in a research institute, while researchers' productivity views improve only with role-specific training and governance measures.
low descriptive openalex
Generative KI in der Wissensarbeit: Wahrnehmung, Nutzen und Akzeptanz von Microsoft 365 Copilot · Carsten F. Schmidt, Sophie Petzolt, Wolfgang Beinhauer, Ingo Weber, Stefan Lange... · Apr 23, 2026
Generative AI can erode its own value: individually rational use drives model collapse that lowers social welfare, with habits spreading harmless-seeming use into high‑value tasks and magnifying damage.
n/a theoretical openalex
Generative artificial intelligence reduces social welfare through model collapse · Fabian Baumann, Erol Akçay, Joshua B. Plotkin · Apr 23, 2026
A four-pillar governance framework (Accountability, Transparency, Fairness, Compliance) offers a practical route for fintechs and regulators to curb algorithmic bias in SME lending; expert validators gave the model strong relevance (mean 4.6/5), though the framework still needs empirical field testing to confirm impacts on inclusion and credit outcomes.
medium review_meta openalex
Corporate-Governance-Driven Algorithmic Fairness in SME Fintech Lending: A Systematic Literature Review with Expert Validation · Chloe Victoria, Daniel Müller · Apr 23, 2026
A two‑level 'harness evolution' framework promises to eliminate manual prompt, tool and orchestration engineering by auto‑evolving agent harnesses and the protocol that evolves them, enabling rapid adaptation of AI agents to new domain workflows.
n/a theoretical arxiv
The Last Harness You'll Ever Build · Haebin Seong, Li Yin, Haoran Zhang · Apr 22, 2026
Value alignment is a governance problem, not just an engineering puzzle: misalignment stems from objective-setting, information flows and whose interests count, so durable solutions require institutions and contested decision‑making rather than only model tweaks.
n/a theoretical arxiv
Relative Principals, Pluralistic Alignment, and the Structural Value Alignment Problem · Travis LaCroix · Apr 22, 2026