The Commonplace
Home Dashboard Papers Evidence Syntheses Digests 🎲

Papers (1139)

Filter: Sort:
Natural-language rendering of a structured prompt protocol (PPS) meaningfully improves LLM alignment and usability—cutting follow-up prompts by about two-thirds—especially for ambiguous business tasks; however, benefits vary by task type and conventional evaluation metrics can mask these practical gains.
medium rct arxiv
Evaluating 5W3H Structured Prompting for Intent Alignment in Human-AI Interaction · Peng Gang · Mar 19, 2026
Institutional design, not model brand, largely determines whether LLM agents misbehave: simulated government agents break rules much more under weak authority structures, and modest safeguards sometimes help but do not reliably prevent serious abuse.
medium quasi_experimental arxiv
I Can't Believe It's Corrupt: Evaluating Corruption in Multi-Agent Governance Systems · Vedanta S P, Ponnurangam Kumaraguru · Mar 19, 2026
A new multi-hop spatial benchmark shows modern vision-language models still falter at compositional spatial reasoning and precise grounding; targeted reinforcement-learning post-training on a dedicated corpus measurably improves both benchmark scores and transfer to embodied manipulation tasks.
medium descriptive arxiv
MultihopSpatial: Multi-hop Compositional Spatial Reasoning Benchmark for Vision-Language Model · Youngwan Lee, Soojin Jang, Yoorhim Cho, Seunghwan Lee, Yong-Ju Lee, Sung Ju Hwan... · Mar 19, 2026
Video filters and avatars erode viewers' trust and confidence but do not make them worse at spotting lies; AI-mediated appearance changes reduce perceived credibility without impairing actual detection of deception.
high rct arxiv
Through the Looking-Glass: AI-Mediated Video Communication Reduces Interpersonal Trust and Confidence in Judgments · Nelson Navajas Fernández, Jeffrey T. Hancock, Maurice Jakesch · Mar 19, 2026
A new Agent Control Protocol codifies cryptographic admission control for enterprise autonomous agents, pairing 36 specification documents with a 22-package Go reference implementation and 51 signed conformance vectors; it aims to let institutions authorize, audit and revoke agent actions without replacing existing RBAC or Zero Trust systems.
n/a descriptive arxiv
Agent Control Protocol: Admission Control for Agent Actions · Marcelo Fernandez · Mar 19, 2026
Prompt framing creates a confirmation-bias blind spot in LLM code reviewers: when changes are framed as bug-free or as security fixes, detection rates fall sharply and attackers can reintroduce known vulnerabilities in up to 88% of trials against an autonomous agent; removing metadata or adding explicit instructions largely undoes the effect.
medium quasi_experimental arxiv
Measuring and Exploiting Confirmation Bias in LLM-Assisted Security Code Review · Dimitris Mitropoulos, Nikolaos Alexopoulos, Georgios Alexopoulos, Diomidis Spine... · Mar 19, 2026
A formal framework separates AI that amplifies human judgment from AI that progressively usurps it, proposing four operational metrics (CAI*, D, HRI, HCDR) to diagnose whether hybrid performance is sustainable; it warns that maximizing short-term capability can undermine long-term human expertise and calls for a cognitive sustainability constraint in system design.
n/a theoretical arxiv
Cognitive Amplification vs Cognitive Delegation in Human-AI Systems: A Metric Framework · Eduardo Di Santi · Mar 19, 2026
Regulatory sandboxes offer a pragmatic route to balance AI innovation and safety by enabling iterative, evidence-based rulemaking; however, their promise hinges on clear boundaries, proportionality, and robust safeguards against capture and rent-seeking.
n/a theoretical semantic_scholar
Experimentalism beyond ex ante regulation: A law and economics perspective on AI regulatory sandboxes · Antonella Zarra · Fetched Mar 19, 2026
Off-the-shelf reasoning AI agents converge to Nash-like behavior in repeated strategic games without post-training alignment, the authors prove and validate in simulations; this suggests many strategic interactions may attain stable equilibria without universal alignment procedures.
medium theoretical arxiv
Reasonably reasoning AI agents can avoid game-theoretic failures in zero-shot, provably · Enoch Hyunwook Kang · Mar 19, 2026
Labeling products or strategy with 'AI' without substance appears to hurt green innovation: Chinese listed firms that AI-wash their annual reports show lower green patenting, a decline transmitted through reputation losses in product markets and increased financing constraints. The effect is strongest for private firms, SMEs and firms in highly competitive sectors, suggesting targeted disclosure and regulatory measures could restore market incentives.
low correlational arxiv
The Spillover Effects of Peer AI Rinsing on Corporate Green Innovation · Li Wenxiu, Wen Zhanjie, Xia Jiechang, Guo Jingqiao · Mar 19, 2026
Zimbabwean banks find AI can broaden access—especially anomaly detection, which explains most measured differences—but rollout is constrained by digital illiteracy, poor internet and high costs, and customers remain wary of opaque credit-scoring and chatbot decisions.
low descriptive openalex
The Impact of Artificial Intelligence on Financial Inclusion in Zimbabwe's Banking Sector: Challenges and Opportunities for Expanding Access to Financial Service · Justin Tinashe Muswaburi, Precious Mandava, Nyasha Mpondwe · Mar 19, 2026
Signalling AI adoption makes employers more attractive to jobseekers; the boost comes from perceptions of greater innovation and is strongest for applicants confident in AI skills.
medium rct openalex
Signaling Organizational Artificial Intelligence Adoption in Recruitment Materials: Role of Perceived Innovation Ability in Organizational Attractiveness · Jialin Cheng, Shunhong Ji · Mar 19, 2026
Financial digital intelligence in China is linked to stronger innovation in strategic emerging industries, with gains driven by improved disclosure, lower transaction costs and deeper industry–university–research ties; effects vary substantially across regions and industry types.
medium correlational openalex
Financial Digital Intelligence and Innovative Development of Strategic Emerging Industries: Impact Effects and Mechanism Tests · Wei XUE, Ziyan LI, Qianyi Lei · Mar 19, 2026
Treat AI governance as a strategic board-level function rather than a technical compliance task: embedding AI oversight in enterprise risk management can reduce deployment risk and speed adoption, turning trustworthy AI practices into a source of competitive advantage.
n/a theoretical openalex
Artificial Intelligence Governance In Corporate Strategy: Ethical Risk, Regulatory Compliance, And Competitive Advantage · Dr. Yemi Martins · Mar 19, 2026
Large language models subtly but systematically change what people mean when they write: heavy LLM users produce nearly 70% more neutral answers and report less creative, less ‘in‑their‑voice’ prose. When asked to revise human essays or write peer reviews, LLMs frequently alter semantics and give reviews that weight clarity/significance less and score papers roughly one point higher on average.
medium quasi_experimental arxiv
How LLMs Distort Our Written Language · Marwa Abdulhai, Isadora White, Yanming Wan, Ibrahim Qureshi, Joel Leibo, Max Kle... · Mar 18, 2026
A test-driven agent workflow that surfaces which tests to run cuts AI-induced regressions by about 70% and raises bug-resolution rates from 24% to 32% on a standard coding benchmark; merely instructing agents to follow TDD increased regressions, while an autonomous improvement loop achieved 60% resolution with zero regressions on a small subset.
medium quasi_experimental arxiv
TDAD: Test-Driven Agentic Development - Reducing Code Regressions in AI Coding Agents via Graph-Based Impact Analysis · Pepe Alonso · Mar 18, 2026
Allowing everyone to game AI decision-support systems can entrench inequality: simultaneous recourse shifts selection thresholds so early winners set the benchmark and direct improvement, amplifying initial disparities and creating persistent performance gaps.
n/a theoretical arxiv
Actionable Recourse in Competitive Environments: A Dynamic Game of Endogenous Selection · Ya-Ting Yang, Quanyan Zhu · Mar 18, 2026
Tax AI only when it threatens cognitive jobs: a theoretical model finds governments should begin taxing AI once its capabilities make cognitive workers consider moving into manual roles, because that threshold marks the point where taxing AI improves labor allocation without unduly distorting capital and growth incentives.
n/a theoretical arxiv
Workers' Incentives and the Optimal Taxation of AI · Jakub Growiec, Klaus Prettner, Maciej Szkróbka · Mar 18, 2026
An AI career coach modestly boosts short-term goal progress compared with no support, but not versus structured written reflection; its chief added value appears to be raising users' sense of accountability, not deeper changes in goal alignment.
medium rct arxiv
AI-Assisted Goal Setting Improves Goal Progress Through Social Accountability · Michel Schimpf, Julian Voigt, Thomas Bohné · Mar 18, 2026
A Governed Memory layer stitches together agent-specific memories into a governed shared store, eliminating cross-entity leakage and enforcement failures while cutting context tokens and maintaining retrieval accuracy; controlled tests and a public benchmark show high recall, strong governance routing, and no quality penalty from schema enforcement.
medium descriptive arxiv
Governed Memory: A Production Architecture for Multi-Agent Workflows · Hamed Taheri · Mar 18, 2026
A multi‑agent LLM sandbox that learns cross‑category consumer preferences and uses mean‑field interactions improves simulated product choice and purchase forecasts, offering a promising platform for scalable market simulation—though its claims rest on simulated environments and depend heavily on the training transaction data.
low descriptive arxiv
MALLES: A Multi-agent LLMs-based Economic Sandbox with Consumer Preference Alignment · Yusen Wu, Yiran Liu, Xiaotie Deng · Mar 18, 2026
Anonymized LLM trading agents produce robust backtest returns (Sharpe ~1.40) and negative controls indicate signals are not just memorized tickers; however, the alpha is sensitive to market regime and limited to short historical windows, raising questions about long-term live performance.
medium quasi_experimental arxiv
Can Blindfolded LLMs Still Trade? An Anonymization-First Framework for Portfolio Optimization · Joohyoung Jeon, Hongchul Lee · Mar 18, 2026
AI is speeding early-stage drug discovery—improving hit-finding, property prediction and protein modelling—but it has not yet delivered a fully AI-originated approved drug; data fragmentation, model limitations and biological uncertainty mean human-led validation and costly clinical testing still determine ultimate success.
medium review_meta semantic_scholar
Has AI Reshaped Drug Discovery, or Is There Still a Long Way to Go? · K. Shree Harini, D. Ezhilarasan · Fetched Mar 18, 2026
AI is speeding up drug discovery and clinical development—cutting timelines and reducing some failure modes—but the economic upside will accrue unevenly because performance depends on proprietary data, heavy upfront costs and regulatory uncertainty.
medium review_meta semantic_scholar
From Algorithm to Medicine: AI in the Discovery and Development of New Drugs · A. Lopes, C. F. Rodrigues, Francisco A. M. Silva · Fetched Mar 18, 2026
AI is already shaving time off early-stage drug discovery but has not moved the needle on Phase II/III success—real value accrues to firms that couple AI with high-quality data, experiments and organizational change.
low review_meta semantic_scholar
Learning from the successes and failures of early artificial intelligence (AI) adoption for drug discovery in Big BioPharma · M. Braddock, Krzysztof Jeziorski · Fetched Mar 18, 2026