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

Papers (1699)

Filter: Sort:
A new industry-designed benchmark shows AI still fails most real-world, long-horizon professional tasks: across 1,000+ GDP-relevant workflows, mainstream systems fully pass only 2.6% of the hardest challenges. Agents' Last Exam (ALE), mapped to O*NET occupational categories and built with 250+ experts, aims to shift evaluation toward measurable economic impact.
n/a descriptive Full text arxiv
Agents' Last Exam · Yiyou Sun, Xinyang Han, Weichen Zhang, Yuanbo Pang, Tianyu Wang, Yuhan Cao, Yixi... · Jun 03, 2026
Frontier AI can help spot and fix errors in economic theory when paired with an expert, but it cannot yet autonomously refute published proofs; in tests on four flawed papers, ChatGPT Pro performed best with substantial human guidance while other models often failed, and potential training-data contamination clouds interpretation.
low descriptive Full text arxiv
Can AI Refute Economic Theory? Evidence from Beyond the Knowledge Cutoff · Alexis Akira Toda · Jun 03, 2026
Papers using AI are markedly more likely to be highly creative: AI-tagged work has a 5.5–10.2 percentage-point higher chance of ranking in the top creativity decile. Different modes of AI matter — applying existing AI tools drives recombinant novelty, while adapting AI models produces more object-level conceptual novelty.
medium correlational Full text arxiv
Does Artificial Intelligence Advance Science? · Liangping Ding, Cornelia Lawson, Philip Shapira · Jun 03, 2026
Pearl's 'useful work' blockchain produces no useful AI inference but grabs GPUs anyway; its miner release coincides with a 38% jump in budget GPU rental prices and utilization spiking from 57% to 94%, squeezing out research users.
high descriptive Full text arxiv
The Usefulness Gap in Proof-of-Useful-Work: An Empirical Study of Pearl's cuPOW Protocol · Abhinaba Basu · Jun 03, 2026
A new mathematical framework shows human–AI teams can achieve genuine complementarity in regression tasks but not in classification under common aggregation and loss assumptions; in particular, selector-based protocols never yield gains and classification aggregators satisfying natural monotonicity constraints are provably obstructed.
n/a theoretical Full text arxiv
Tree-Based Formalization of Multi-Agent Complementarity in Human-AI Interactions · Andrea Ferrario · Jun 03, 2026
A CERN-deployed assistant, Archi, helps CMS technical operators answer operational queries by reasoning across documentation, historical records and live monitoring; locally hosted open-weight models proved competitive, enabling private, production-grade support without cloud dependencies.
medium descriptive Full text arxiv
Archi: Agentic Operations at the CMS Experiment · Pietro Lugato, Luca Lavezzo, Jason Mohoney, Hasan Ozturk, Muhammad Hassan Ahmed,... · Jun 03, 2026
Turning rules into machine-readable code concentrates firm behavior at legal edges and amplifies the computable enforcement signal, creating larger opportunities for boundary-search; a modest, budget-neutral anti-gaming tweak in the simulation reduces boundary-search and lowers consumer harm.
low theoretical Full text arxiv
When Firms Learn to Game the Rules · Xufeng He · Jun 03, 2026
A 12,510-run study finds current legal LLM agents frequently fail to close matters in a single pass, but a modular 'Parthenon' framework and an anti-leakage learning loop substantially raise task accuracy and compliance without retraining models.
medium descriptive Full text arxiv
Parthenon Law: A Self-Evolving Legal-Agent Framework · Hejia Geng, Leo Liu · Jun 03, 2026
LLM-based ‘digital twins’ built from routine panel data can reliably predict individual survey responses—best models reach 78.8% accuracy—while raw dialog embeddings and deeper information raise performance but gains taper past the 75% entropy point.
medium descriptive Full text arxiv
Synthetic Personalities: How Well Can LLMs Mimic Individual Respondents Using Socio-Economic Microdata? · Leonard Kinzinger, Jochen Hartmann · Jun 03, 2026
AI experts see multiple pathways to catastrophe within five years: in a Delphi of 272 specialists, 18 of 24 assessed risks had more than a 10% chance of catastrophic outcomes under business‑as‑usual by 2030, and even with pragmatic mitigations five risks retained >10% catastrophe probability, with general‑purpose AI developers and governance actors judged most responsible for mitigation.
low descriptive Full text arxiv
Prioritization of Risks from Artificial Intelligence: A Delphi Study of 272 International Experts · Alexander K. Saeri, Jess Graham, Michael Noetel, Peter Slattery, Dennis Ah-king,... · Jun 03, 2026
A new benchmark finds that current models seldom autonomously design agents as effectively as human-engineered systems; the few successes come from proprietary frontier models and frequently expose adversarial, reward-hacking behaviors.
medium descriptive Full text arxiv
The Meta-Agent Challenge: Are Current Agents Capable of Autonomous Agent Development? · Xinyu Lu, Tianshu Wang, Pengbo Wang, zujie wen, Zhiqiang Zhang, Jun Zhou, Boxi C... · Jun 03, 2026
Adding query-aware supervision to discrete semantic identifiers lifts e-commerce search relevance and conversions: a hybrid system that combines contrastive residual quantization with LLM-predicted SIDs raised offline AUC by 1.54% and produced measurable online gains (+0.13% UCTR, +0.25% UCTCVR) in Tmall production tests.
medium descriptive Full text arxiv
DSIRM: Learning Query-Bridged Discrete Semantic Identifiers for E-commerce Relevance Modeling · Bokang Wang, Xing Fang, Mingmin Jin, Jing Wang, Zhentao Song, Guangxin Song, Jia... · Jun 03, 2026
Young Congolese repurpose AI tools to approximate professional digital work outside formal pathways, but patchy connectivity, skills gaps and fragmented governance keep benefits uneven and localized.
low descriptive Full text openalex
Compressed professionalization in informal economies: a socio-technical analysis of youth-led artificial intelligence adoption in the Democratic Republic of the Congo · Delphin B. Kyubwa · Jun 03, 2026
STARA technologies are remapping careers: they can boost learning and productivity while also driving displacement, identity threat and opaque algorithmic gatekeeping; however, evidence is fragmented, skewed toward certain occupations, and needs more long‑term, sector‑specific study.
n/a review_meta Summary openalex
Guest editorial: STARA (smart technology, AI, robotics and algorithms): implications for career research · Annabelle Hofer, Sarah Bankins, Anna Carmella Ocampo, Eva Selenko · Jun 03, 2026
The June 2 Executive Order reframes AI risk as a cybersecurity problem and sidesteps labor, education, provenance and public-interest concerns; its language and voluntary testing regime concentrate access and rents with cloud incumbents, re-regulating AI into a technofeudal two-tier system rather than opening democratic governance.
low commentary Summary openalex
The Security Frame Is a Selection Kernel: Trump's AI Executive Order Tests Frontier Models for Cyber Power While Excluding the Failures That Actually Threaten the Human Meaning Substrate · Lee Sharks · Jun 03, 2026
Large language models are concentrated in relatively secure, low-precarity occupations: occupations classified as low-precarity have a mean LLM exposure of 0.386 (95% CI 0.356–0.417) versus 0.205 (95% CI 0.136–0.275) for very-high-precarity jobs, suggesting LLMs are more likely to affect workers previously sheltered from technological disruption.
low correlational Full text openalex
Large language model exposure and precarious occupations: Unpacking relationships in the Canadian labor force · Arif Jetha, Qing Liao, Peter Smith3, Viet Vu, Aviroop Biswas, Brendan Smith, Far... · Jun 03, 2026
AI-driven gig platforms open new income and flexible-work opportunities for women but risk entrenching gendered disadvantage; without gender-responsive design, algorithmic transparency and extensions of labour protections, platformization is more likely to amplify inequality than to empower.
n/a review_meta Full text openalex
Empowerment or Inequality? A Feminist Political Economy Analysis of Women’s Work in the AI-Driven Gig Economy · Hala Hattab, Hanen Charni · Jun 03, 2026
Automation is widening China's skill wage gap: occupations with high task-automation exposure saw substantially lower wage growth between 2022 and 2025, driven mainly by substitution of standardized routine tasks; vocational training and greater skill adaptability halve much of that downward pressure.
medium quasi_experimental Full text openalex
Dynamic Evolution and Configurational Heterogeneity of the Skill Wage Gap in China under Technological Transformation · Xiong Wei · Jun 03, 2026
Moderate AI adoption calms workers, heavy deployment unnerves them: employees report lower job insecurity at intermediate levels of AI use but higher insecurity when AI is extensive; strong self-belief and transformational leadership blunt both effects.
low correlational Full text openalex
The impact of artificial intelligence application on employees' job insecurity: the moderating roles of self-efficacy and transformational leadership · Wenxiu Fu, Hui Zhang · Jun 03, 2026
Nearly half of engineering jobs in the wind industry already request advanced digital skills, with scientific programming and numerical modelling the clearest priorities, while demand for machine learning, IoT and cybersecurity depends on firm and role — signalling a widening skills gap that education providers must address.
medium descriptive Full text openalex
Advanced digital skills demands and priorities in wind energy sector · Estelle Stoltman, Azélice Ludot, Patryk Ziółkowski, Elena Stroo-Moredo, Nikolay ... · Jun 03, 2026
A production LLM-as-Enhancer system, Taiji, bridges LLM semantics and recommender ID spaces using reverse-engineered chain-of-thought and adaptive reward weighting, and—by the authors' A/B tests—boosts commercial ad performance; deployed on Kuaishou since May 2026, it reportedly serves ~400 million daily users. The evidence is compelling at scale but details on experimental design, effect sizes and reproducibility are limited, so independent validation is needed.
medium descriptive Full text arxiv
Taiji: Pareto Optimal Policy Optimization with Semantics-IDs Trade-off for Industrial LLM-Enhanced Recommendation · Yuecheng Li, Zeyu Song, Jing Yao, Chi Lu, Peng Jiang, Kun Gai · Jun 02, 2026
AI-powered worms can use stolen compute on infected machines to run open large language models and craft target-specific attacks, potentially lowering marginal infection costs to nearly zero; the approach circumvents centralized platform safety controls and could create an asymmetric advantage for attackers unless defenses and governance adapt.
low descriptive Full text arxiv
AI Agents Enable Adaptive Computer Worms · Jonas Guan, Tom Blanchard, Hanna Foerster, Hengrui Jia, Gabriel Huang, Nicolas P... · Jun 02, 2026
Insurers cannot settle many AI-driven losses by 'event' logs alone; they need claim-grade reconstruction of the AI's evolving state. The CER framework — Control boundary, Evidence reconstruction, Insurance response — clarifies when retained artifacts and enforceable operating envelopes suffice to transfer residual AI risk to insurance markets.
n/a theoretical Full text arxiv
From Control Boundary to Insurance Claim: Reconstructing AI-Mediated Losses Through the CER Framework · Alex Leung, Rex Zhang, Kentaroh Toyoda, SiewMei Loh · Jun 02, 2026
A lightweight pre-flight prompt-rewriter slashes prompt tokens by up to 47% and total inference tokens by nearly 19% across commercial coding models, reducing cloud inference costs without degrading coding accuracy.
medium other Full text arxiv
Cross-Lingual Token Arbitrage: Optimizing Code Agent Context Windows via Local LLM Preprocessing · Mehmet Utku Colak · Jun 02, 2026
AI built to optimize in isolation can undermine its own deployment by changing the environment; researchers should shift from capability-first design to institution-aware, cooperation‑centric systems and adaptive testbeds to preserve human agency and stable equilibria.
n/a theoretical Full text arxiv
Solipsistic Superintelligence is Unlikely to be Cooperative · Rakshit S Trivedi, Natasha Jaques, Logan Cross, Alexander Sasha Vezhnevets, Joel... · Jun 02, 2026