The Commonplace
Home Dashboard Papers Evidence Digests 🎲

Papers (639)

Filter: Sort:
Generative quantum-code models claim syntactic and semantic competence, but none demonstrate real-device execution, leaving practical deployability and commercial value uncertain; investors and researchers should treat reported performance as provisional until hardware-level validation is provided.
medium review_meta arxiv
Generative AI for Quantum Circuits and Quantum Code: A Technical Review and Taxonomy · Juhani Merilehto · Mar 17, 2026
A modular, multi-agent interviewing system outperforms single‑agent LLM assessments on robustness and auditability, delivering high accuracy and candidate satisfaction while reducing subjective bias; its gains hinge on rubric design and come with higher engineering and compute costs.
medium descriptive arxiv
CoMAI: A Collaborative Multi-Agent Framework for Robust and Equitable Interview Evaluation · Gengxin Sun, Ruihao Yu, Liangyi Yin, Yunqi Yang, Bin Zhang, Zhiwei Xu · Mar 17, 2026
Leaderboard accuracy for frontier LLMs is materially inflated: about 13.8% of MMLU items show evidence of training-data exposure, boosting reported scores (particularly in STEM/professional subjects) and creating memorization fingerprints that can mislead buyers, investors and benchmark-driven procurement.
medium descriptive arxiv
Are Large Language Models Truly Smarter Than Humans? · Eshwar Reddy M, Sourav Karmakar · Mar 17, 2026
A modular pre-processor that recovers implied premises, weights recent context, and verifies semantic alignment can cut misaligned or off-topic AI replies and reduce correction cycles; the approach promises productivity gains and changed labor demands but entails extra compute, latency and careful tuning.
n/a theoretical arxiv
A Context Alignment Pre-processor for Enhancing the Coherence of Human-LLM Dialog · Ding Wei · Mar 17, 2026
Algorithmic lenders improve repayment rates and borrower resilience but raise financial stress; stronger borrower skills and governance arrangements convert algorithmic lending from a mixed risk into net benefit.
low correlational openalex
Architecting financial well-being in algorithmic credit systems: The roles of human capability and institutional design · Christian Anthony Flores · Mar 17, 2026
Moderate data sharing maximizes welfare: a theoretical model finds that sharing big data incurs privacy costs that can slow welfare initially but, when calibrated optimally, drives long-run growth through innovation and multiplier effects. Policymakers should balance privacy protections against the long-term productivity gains of data-driven R&D.
n/a theoretical openalex
Study on the impact of big data sharing on individuals’ welfare—from the perspective of consumption and privacy · Hanghang Dong, Xiaoming Li, Yadi Liu, Chan Wang · Mar 17, 2026
Startups report productivity and skills gains from AI: interviews with a dozen employees find faster task completion, better decision-making and rising digital competencies, though initial adaptation hurdles persist.
low descriptive openalex
AI-AUGMENTED WORKFORCE: THE IMPACT OF ARTIFICIAL INTELLIGENCE INTEGRATION ON EMPLOYEE COMPETENCIES AND PERFORMANCE IN THE DIGITAL ERA · Kadar Nurjaman, Avid Leonardo Sari, A Syathir Sofyan, Rahayu Kusumadewi, Nurhaet... · Mar 17, 2026
Generative AI on Microsoft Azure can yield positive enterprise ROI—largely through productivity and cost efficiencies—provided firms align deployments with business processes, Azure OpenAI/Azure ML integration and strict FinOps cost governance; integration reduces total cost of ownership and improves scalability and compliance.
low descriptive openalex
Measuring Business ROI of Generative AI Adoption on Azure Cloud Platforms · Rahul Modi · Mar 17, 2026
AI can cut material waste and boost factory resource efficiency—case evidence shows up to 25% efficiency gains and 30% less scrap—but gains are uneven, with SMEs and firms in emerging markets falling behind due to data and capability constraints.
medium review_meta openalex
Artificial intelligence as a catalyst for the circular economy: a mixed-methods analysis of sustainable production ecosystems (2023–2024) · A. Lopez, Carlos Luengo Vera, Ángel Javier Álvarez Miguel · Mar 17, 2026
Automated tax systems often raise compliance and cut administration costs, but their benefits are conditional: low digital literacy, weak trust and unclear regulation erode gains and can trigger taxpayer resistance.
medium review_meta openalex
The Influence of Automation on Tax Compliance Behaviour · Alexander Oluka · Mar 17, 2026
Marrying formal argumentation with large language models could make AI decisions inspectable and contestable, creating demand for verifiable AI services in regulated sectors; at the same time it will shift experts toward adjudication and oversight and create new governance needs.
n/a theoretical arxiv
Argumentative Human-AI Decision-Making: Toward AI Agents That Reason With Us, Not For Us · Stylianos Loukas Vasileiou, Antonio Rago, Francesca Toni, William Yeoh · Mar 16, 2026
Generative legal AI delivers fluent but fallible advice: its persuasive hallucinations and opaque reasoning risk embedding errors into legal processes and eroding judicial independence unless regulators demand effective human oversight and verifiable verification, a shift that will reshape adoption, liability costs and legal labor markets.
n/a commentary arxiv
Why Avoid Generative Legal AI Systems? Hallucination, Overreliance, and their Impact on Explainability · Gizem Gültekin Varkonyi · Mar 16, 2026
An AI pipeline produced a machine-verified mathematical formalization in 10 days under a single supervisor, automating coding and lemma-proving but still requiring expert oversight; the case shows high-skill scientific tasks can be substantially automated in practice, though verification and definition-alignment remain critical bottlenecks.
low descriptive arxiv
Semi-Autonomous Formalization of the Vlasov-Maxwell-Landau Equilibrium · Vasily Ilin · Mar 16, 2026
The architecture of long‑lived, physical AI agents will determine whether they deliver broad economic gains or become costly, unsafe liabilities. Prioritizing evolvability, secure identities, semantic interoperability and observability can prevent expensive lock‑in, reduce systemic risk, and unlock faster, safer adoption.
n/a theoretical arxiv
The Internet of Physical AI Agents: Interoperability, Longevity, and the Cost of Getting It Wrong · Roberto Morabito, Mallik Tatipamula · Mar 16, 2026
Emboldened autonomy can amplify crises: embodied AI in critical infrastructure must be constrained and paired with human oversight to prevent cascading failures, with clear, auditable allocations of machine capability and human judgement guiding deployment and regulation.
low theoretical arxiv
Resilience Meets Autonomy: Governing Embodied AI in Critical Infrastructure · Puneet Sharma, Christer Henrik Pursiainen · Mar 16, 2026
People cooperate readily with a powerful chatbot, but not as completely as with humans: natural-language chat with GPT-5.2 yields high initial cooperation yet a persistent plateau below the near-universal cooperation seen in human pairs, because conversations with the AI prompt explicit conditional rules rather than trust-building social signals.
medium rct arxiv
Playing Against the Machine: Cooperation, Communication, and Strategy Heterogeneity in Repeated Prisoner's Dilemma · Chowdhury Mohammad Sakib Anwar, Konstantinos Georgalos · Mar 16, 2026
A single compromised LLM can hijack group decisions in multi-agent systems via a persuasion cascade; dynamically reducing trust in suspected adversaries (while tuning agent stubbornness and scale) curbs influence at the cost of some coordination or compute overhead.
low theoretical arxiv
Don't Trust Stubborn Neighbors: A Security Framework for Agentic Networks · Samira Abedini, Sina Mavali, Lea Schönherr, Martin Pawelczyk, Rebekka Burkholz · Mar 16, 2026
A new Pokemon-based benchmark exposes large capability gaps on multi-agent, partial-observability and long-horizon planning tasks: specialist RL systems and human experts outperform generalist LLMs, and a 20M+ trajectory dataset plus a NeurIPS competition confirm strong community interest and reproducible evaluation.
n/a descriptive arxiv
The PokeAgent Challenge: Competitive and Long-Context Learning at Scale · Seth Karten, Jake Grigsby, Tersoo Upaa, Junik Bae, Seonghun Hong, Hyunyoung Jeon... · Mar 16, 2026
iDaVIE—an open-source VR toolkit—lets astronomers interactively inspect and edit very large 3D spectral data cubes in real time, accelerating quality-control and mask-refinement workflows and producing higher-quality inputs for ML pipelines; adoption improves inspection throughput in practice but requires VR-capable hardware and integration effort.
low descriptive arxiv
iDaVIE v1.0: A virtual reality tool for interactive analysis of astronomical data cubes · Alexander Sivitilli, Lucia Marchetti, Angus Comrie, P. Cilliers Pretorius, Thijs... · Mar 16, 2026
Prepackaged 'skills' for coding agents rarely move the needle: in a 565-task benchmark across 49 skills, most skills produced no test-pass improvements and only a handful delivered substantial gains, with some even harming outcomes when guidance conflicted with project context.
medium quasi_experimental arxiv
SWE-Skills-Bench: Do Agent Skills Actually Help in Real-World Software Engineering? · Tingxu Han, Yi Zhang, Wei Song, Chunrong Fang, Zhenyu Chen, Youcheng Sun, Lijie ... · Mar 16, 2026
Current AI architectures lack the mechanisms for sustained, autonomous learning; a three-part design that integrates observation, active experimentation and an internal meta-controller could unlock more adaptable, sample-efficient agents and accelerate automation in embodied and social tasks.
n/a theoretical arxiv
Why AI systems don't learn and what to do about it: Lessons on autonomous learning from cognitive science · Emmanuel Dupoux, Yann LeCun, Jitendra Malik · Mar 16, 2026
Structured domain guidance (SKILL.md) reliably boosts LLM agents' telecom operations competence, raising success rates across models by up to ~19 percentage points; MiniMax M2.5 leads at 81.1% success with the skill augmentation.
medium descriptive arxiv
SKILLS: Structured Knowledge Injection for LLM-Driven Telecommunications Operations · Ivo Brett · Mar 16, 2026
AI assistant Gemini did not improve code security in a controlled developer experiment; developers' own programming experience—rather than the tool—drove more secure outcomes.
medium rct arxiv
The Impact of AI-Assisted Development on Software Security: A Study of Gemini and Developer Experience · Nadine Jost, Benjamin Berens, Manuel Karl, Stefan Albert Horstmann, Martin Johns... · Mar 16, 2026
A probe-then-plan LLM system deployed at JD.com lifts relevant recall and conversion, producing measurable increases in gross merchandise value; lightweight retrieval probes let the planner ground decisions in live inventory without incurring prohibitive latency.
medium rct arxiv
Probe-then-Plan: Environment-Aware Planning for Industrial E-commerce Search · Mengxiang Chen, Zhouwei Zhai, Jin Li · Mar 16, 2026
Tailored AI coaching measurably improves how people express empathy: a randomized trial finds personalized LLM feedback produces more normatively empathic replies than generic training or no intervention. But identical messages lose perceived authenticity when recipients are told they came from AI, highlighting a trade-off between content quality and attribution.
high rct arxiv
Practicing with Language Models Cultivates Human Empathic Communication · Aakriti Kumar, Nalin Poungpeth, Diyi Yang, Bruce Lambert, Matthew Groh · Mar 16, 2026