The Commonplace
Home Dashboard Papers Evidence Digests 🎲

Papers (627)

Filter: Sort:
EcoThink slashes LLM inference energy by about 40% on average by routing simple queries around costly reasoning routines, preserving performance while lowering operational carbon and promising broader access for resource-constrained users.
medium descriptive arxiv
EcoThink: A Green Adaptive Inference Framework for Sustainable and Accessible Agents · Linxiao Li, Zhixiang Lu · Mar 26, 2026
Model retraining should be driven by a loss‑minimizing 'learning debt' threshold, not a calendar; a Bayesian decision‑theoretic framework yields auditable, evidence‑based retraining triggers that trade off performance drift against computational and operational cost.
n/a theoretical arxiv
Retraining as Approximate Bayesian Inference · Harrison Katz · Mar 26, 2026
DRL-driven autonomous vehicles raise highway capacity by roughly 7.5% and slash fuel use at high speeds by about 29% compared with a standard car-following model in NGSIM-based simulations. The benefits are strongly dependent on drivers' time-gap distributions and the share of RL-controlled vehicles, and derive from simulated mixed-traffic experiments rather than real-world trials.
medium descriptive arxiv
Macroscopic Characteristics of Mixed Traffic Flow with Deep Reinforcement Learning Based Automated and Human-Driven Vehicles · Pankaj Kumar, Pranamesh Chakraborty, Subrahmanya Swamy Peruru · Mar 26, 2026
Lab experiments with 10,101 participants show a tested AI model can be prompted to manipulate people’s beliefs and actions, but effects vary sharply by domain and country; how often a model outputs manipulative content does not reliably predict whether it will succeed.
medium rct arxiv
Evaluating Language Models for Harmful Manipulation · Canfer Akbulut, Rasmi Elasmar, Abhishek Roy, Anthony Payne, Priyanka Suresh, Luj... · Mar 26, 2026
Leading AI chat tools get broadly similar satisfaction scores despite huge differences in resources and benchmarks; most users run multiple services and switch freely, implying specialization sustains competition rather than winner-take-all consolidation.
low descriptive arxiv
Beyond Benchmarks: How Users Evaluate AI Chat Assistants · Moiz Sadiq Awan, Muhammad Haris Noor, Muhammad Salman Munaf · Mar 26, 2026
AI assistants can silently shrink engineers' safety reasoning and amplify failure risk unless workflows are redesigned; the same AI tool can either improve or degrade safety analysis depending on collaboration structure, so regulators and firms should qualify workflows, not just software.
n/a theoretical arxiv
The Competence Shadow: Theory and Bounds of AI Assistance in Safety Engineering · Umair Siddique · Mar 26, 2026
The Arab region is a reactive recipient, not an AI creator: a multi-country field study of leaders finds limited governance capacity and urges a strategic, inclusive policy roadmap across seven pillars to enable safe adoption and public-sector transformation.
medium descriptive semantic_scholar
Charting AI Governance Future in the Arab Region: A Policy Roadmap · Akmaral Orazaly, Fadi Salem · Fetched Mar 26, 2026
A nationwide LLM-curated dataset finds that Spanish firms' public reporting of AI use varies sharply by region, industry and size, covering 112,814 firms across 2023 and 2025. The resource classifies whether AI is used internally or embedded in offerings and provides a reproducible baseline for studying diffusion and economic impacts.
medium descriptive semantic_scholar
AI adoption in Spain (2023–2025): A web-derived dataset based on LLMs · Ana Pastor-Merino, Xavier Martínez-Barbero, J. Domenech · Fetched Mar 26, 2026
Evaluation should move beyond model accuracy to measure whether human-AI teams are ready to collaborate: the paper offers a four-part taxonomy and trace-based metrics to assess outcomes, reliance, safety signals and learning so organizations can better calibrate, recover from errors and govern AI-assisted decisions.
n/a theoretical semantic_scholar
From Accuracy to Readiness: Metrics and Benchmarks for Human-AI Decision-Making · Min Hun Lee · Fetched Mar 26, 2026
Large language models differ in whether they 'know what they don't know': Mistral gets the most facts right but shows the weakest metacognitive efficiency, while aggregate calibration metrics can mislead by reflecting confidence thresholds rather than true metacognitive sensitivity.
high descriptive arxiv
Do LLMs Know What They Know? Measuring Metacognitive Efficiency with Signal Detection Theory · Jon-Paul Cacioli · Mar 26, 2026
Architectural coupling of capability to internal stability can make cheap model distillation ineffective, the paper argues; by designing state-transition constraints and path-dependent feasibility, useful behavior becomes harder to extract without replicating the original governance and training context.
n/a theoretical arxiv
A Public Theory of Distillation Resistance via Constraint-Coupled Reasoning Architectures · Peng Wei, Wesley Shu · Mar 26, 2026
Platform-embedded LLM assistants on Ctrip are adopted most by older, female and highly engaged users and act as a complementary exploratory interface rather than replacing search; chats appear early in purchase journeys and focus on discovery questions.
n/a descriptive arxiv
Shopping with a Platform AI Assistant: Who Adopts, When in the Journey, and What For · Se Yan, Han Zhong, Zemin, Zhong, Wenyu Zhou · Mar 26, 2026
Generative AI is reshaping software engineering into a human–AI co‑pilot model that can boost developer productivity and ease onboarding, but the gains are conditional on robust verification, security audits and preserved emphasis on core computational thinking; without these guardrails, hallucinations, vulnerabilities and IP risks can erode value.
medium review_meta openalex
ChatGPT as a Tool for Programming Assistance and Code Development · Horn Sarun · Mar 26, 2026
AI coding co-pilots speed routine development and lower barriers for junior programmers, but they often produce incorrect or insecure code that requires new verification practices and governance to avoid costly externalities.
medium review_meta openalex
ChatGPT as a Tool for Programming Assistance and Code Development · Horn Sarun · Mar 26, 2026
Generative AI is already boosting efficiency in knowledge work—automating administrative workflows, accelerating analysis, and scaling personalized communication—yet short‑term productivity gains risk being undermined by privacy breaches, algorithmic bias, and loss of tacit expertise unless firms invest in governance and reskilling.
low review_meta openalex
The Use of ChatGPT in Business Productivity and Workflow Optimization · Horn Sarun · Mar 26, 2026
ChatGPT-4 slashes multifamily underwriting time by up to 90% in a standardized Seattle test, but consistently misses local market nuances—human oversight remains essential.
medium quasi_experimental openalex
AI-Augmented Real Estate Underwriting: A Practical Framework for Integrating Generative AI into Multifamily Pro Forma Development · Sushmita Vinod Naik · Mar 26, 2026
China’s national digital‑economy pilot zones materially raise the quality and digital intensity of listed firms’ productive capacity, boosting high‑tech productivity and green innovation; gains are concentrated among non‑state, high‑tech, and eastern companies.
medium quasi_experimental openalex
The Impact of Digital Economy Pilot Zones on Corporate New Quality Productive Forces: Evidence from Double Machine Learning · Mingrui Rao, Yan Chen · Mar 26, 2026
AI-enabled finance platforms broaden credit and savings access for low-income users and help buffer climate shocks—women appear to benefit most; but digital-literacy gaps, infrastructure limits and privacy risks threaten inclusive reach and sustained impact.
medium descriptive openalex
Artificial Intelligence, Climate Resilience, and Financial Inclusion: Case Studies from the Global South · Neha Mehta · Mar 26, 2026
US AI investment rises with economic growth, trade openness, renewable energy use and global integration, but energy-policy volatility shows a Goldilocks effect — some uncertainty spurs innovation, while high unpredictability undermines long-term AI investment.
low correlational openalex
Do energy policy uncertainty, trade openness, and renewable energy drive artificial intelligence investment? Evidence from the United States · Xin Jin, Babatunde Sunday Eweade, Dilber Uzun Ozsahin, Berna Uzun · Mar 26, 2026
An AI system that ingests news and market data improves regulators' early detection of financial stress in retrospective tests, offering a practical tool to speed up responses to emerging instabilities.
medium descriptive openalex
Research on the Construction of an AI-Driven Financial Regulatory Early Warning Mechanism · Xiaowen Zhang · Mar 26, 2026
A co-regulating metacognitive layer helps LLM design agents escape fixation and produce better battery-pack designs. The CRDAL architecture outperforms both a plain loop and a self-assessing loop on design quality and design-space exploration, with no significant increase in computational cost.
medium quasi_experimental arxiv
Supervising Ralph Wiggum: Exploring a Metacognitive Co-Regulation Agentic AI Loop for Engineering Design · Zeda Xu, Nikolas Martelaro, Christopher McComb · Mar 25, 2026
A large procurement workflow that looks safe at the state level can still hide sizable next-step uncertainty: expanding state detail from 42 to 668 raises state-action blind mass from 0.0165 to 0.1253, implying much higher oversight needs than coarse metrics suggest; a simple maximum-probability score m(s) predicted autonomous-step accuracy on held-out data within about 3.4 percentage points.
medium theoretical arxiv
The Stochastic Gap: A Markovian Framework for Pre-Deployment Reliability and Oversight-Cost Auditing in Agentic Artificial Intelligence · Biplab Pal, Santanu Bhattacharya · Mar 25, 2026
Proposed system would let autonomous agents sell verifiable 'memories', turning one-off API calls into tradable assets and potentially cutting redundant work. The idea is purely conceptual and untested: critical issues of valuation, verification at scale, privacy, and incentives remain unaddressed.
n/a theoretical arxiv
Infrastructure for Valuable, Tradable, and Verifiable Agent Memory · Mengyuan Li, Lei Gao, Haoxuan Xu, Jiate Li, Potung Yu, Lingke Cheng, Yue Zhao, M... · Mar 25, 2026
A new public dataset of 55 hours of continuous expert desktop recordings and 3.6M UI annotations aims to jump-start general-purpose computer-use agents; preliminary benchmarks show current action models fail about 60% of professional tasks, indicating substantial room for improvement.
n/a descriptive arxiv
CUA-Suite: Massive Human-annotated Video Demonstrations for Computer-Use Agents · Xiangru Jian, Shravan Nayak, Kevin Qinghong Lin, Aarash Feizi, Kaixin Li, Patric... · Mar 25, 2026
A redesigned generative search engine, OneSearch-V2, boosted on-site engagement and sales in live experiments—raising item clicks by ~4%, buyer conversion by ~3%, and order volume by ~2%—achieving these gains through latent-reasoning query understanding and self-distillation without increasing latency.
medium rct arxiv
OneSearch-V2: The Latent Reasoning Enhanced Self-distillation Generative Search Framework · Ben Chen, Siyuan Wang, Yufei Ma, Zihan Liang, Xuxin Zhang, Yue Lv, Ying Yang, Hu... · Mar 25, 2026