Evidence (8974 claims)
Search and filter individual claims pulled from the papers. Looking for a specific finding ("what's the effect on wages?"), you're in the right place. Want to compare whole outcome categories against each other instead? Use the Evidence Explorer.
The board below groups claims two ways: by broad theme (nine paper-level topics) and by outcome category (the 34 claim-level outcomes that the Explorer and Syntheses also use).
Browse by theme
Nine broad, paper-level topics. Click one to filter the claims below.
Adoption
10085 claims
Filter claims →
Productivity
8974 claims
Filtered →
Governance
8062 claims
Filter claims →
Human-AI Collaboration
7749 claims
Filter claims →
Org Design
5057 claims
Filter claims →
Innovation
4896 claims
Filter claims →
Labor Markets
4088 claims
Filter claims →
Skills & Training
3372 claims
Filter claims →
Inequality
2377 claims
Filter claims →
Claims by outcome category
Counts by direction of finding. These are the same 34 outcome categories the Explorer compares and the Syntheses are written for. A linked row has a published synthesis.
| Outcome | Positive | Negative | Mixed | Null | Total |
|---|---|---|---|---|---|
| Other | 882 | 244 | 117 | 1097 | 2424 |
| Governance & Regulation | 1010 | 469 | 229 | 135 | 1875 |
| Organizational Efficiency | 977 | 235 | 149 | 90 | 1462 |
| Technology Adoption Rate | 781 | 299 | 143 | 128 | 1362 |
| Research Productivity | 506 | 155 | 74 | 363 | 1110 |
| Output Quality | 555 | 219 | 71 | 70 | 915 |
| Decision Quality | 395 | 200 | 95 | 54 | 751 |
| Firm Productivity | 523 | 67 | 101 | 27 | 724 |
| AI Safety & Ethics | 262 | 309 | 75 | 36 | 688 |
| Market Structure | 195 | 201 | 135 | 30 | 566 |
| Task Allocation | 248 | 77 | 96 | 38 | 464 |
| Innovation Output | 300 | 34 | 55 | 20 | 411 |
| Skill Acquisition | 207 | 75 | 65 | 21 | 368 |
| Employment Level | 138 | 67 | 119 | 24 | 350 |
| Fiscal & Macroeconomic | 156 | 80 | 53 | 33 | 329 |
| Task Completion Time | 211 | 38 | 13 | 16 | 280 |
| Firm Revenue | 183 | 52 | 29 | 5 | 270 |
| Consumer Welfare | 131 | 77 | 48 | 13 | 269 |
| Inequality Measures | 50 | 141 | 54 | 9 | 254 |
| Worker Satisfaction | 104 | 85 | 25 | 13 | 227 |
| Error Rate | 87 | 112 | 11 | 5 | 215 |
| Automation Exposure | 69 | 69 | 37 | 20 | 198 |
| Wages & Compensation | 102 | 49 | 31 | 11 | 193 |
| Team Performance | 115 | 30 | 30 | 11 | 187 |
| Regulatory Compliance | 88 | 74 | 17 | 7 | 186 |
| Training Effectiveness | 109 | 22 | 14 | 21 | 168 |
| Developer Productivity | 116 | 21 | 15 | 8 | 161 |
| Job Displacement | 12 | 92 | 26 | 1 | 131 |
| Hiring & Recruitment | 57 | 12 | 9 | 5 | 83 |
| Skill Obsolescence | 6 | 59 | 10 | 2 | 77 |
| Social Protection | 43 | 17 | 8 | 2 | 70 |
| Creative Output | 35 | 21 | 9 | 4 | 70 |
| Labor Share of Income | 18 | 23 | 17 | 1 | 59 |
| Worker Turnover | 15 | 16 | — | 4 | 35 |
| Industry | — | — | — | 1 | 1 |
Productivity
Remove filter
To mitigate data snooping biases, the closed-loop system imposes strict empirical discipline through out-of-sample validation and economic rationale requirements.
Description of model validation protocol in the paper (use of out-of-sample validation and economic rationale filters); supports claim that these steps are used to reduce data-snooping risk.
The approach operationalizes the model as a self-directed engine that endogenously formulates interpretable trading signals (rather than relying on sequential manual prompts).
Methodological description and implementation details in the paper describing how the model generates signals autonomously and interpretable outputs; empirical example applied to U.S. equity market is referenced to illustrate operation.
We develop an autonomous framework for systematic factor investing via agentic AI.
Statement of methodological contribution in the paper (framework description); no sample size or empirical test required for the descriptive claim.
The technology particularly benefits less experienced practitioners by providing comprehensive starting points for legal research, while experienced attorneys can use it for quality control and initial drafts.
Authors' interpretation of AI outputs from the experiment and reasoning about how those outputs map onto different practitioner needs (qualitative judgment).
The analysis reveals AI’s potential to transform law firm economics by dramatically reducing research time while maintaining analytical quality, though careful attorney oversight remains essential.
Inference from the experimental finding that four AI systems produced substantive analysis comparable to junior-associate work on one transcript and the stated observation about traditional research time (8–40 hours); authors' qualitative judgment about economic implications and need for oversight.
Statutory and regulatory citations proved generally accurate and useful.
Authors' examination of statutory and regulatory references produced by the four AI engines in the experiment, judged to be generally correct and helpful.
All four engines successfully spotted legal issues, assessed claim strengths and weaknesses, and suggested follow-up investigation—tasks that traditionally required eight to forty hours of junior attorney research time.
Observed outputs from the four AI engines on the single transcript showing issue-spotting, strengths/weaknesses assessment, and suggested follow-ups; comparison to typical junior attorney research time (stated as 8–40 hours).
Contemporary generative AI performs sophisticated legal analysis comparable to experienced associates, correctly identifying major employment law claims including ADA violations, Title VII discrimination, OSHA retaliation, FMLA interference, and workers’ compensation retaliation.
Qualitative assessment of outputs from the four AI engines applied to the single hypothetical transcript; comparison against expected legal claims (authors' judgment that outputs matched those an experienced associate would produce).
Four major generative AI engines—DeepSeek, Claude, ChatGPT, and Grok—are useful legal analysis tools for employment law practitioners.
Experimental evaluation in which a single hypothetical client interview transcript was submitted to each of the four AI systems and their outputs were assessed by the authors.
Policy recommendations: increase investment in AI research and expansion; promote AI-driven robotics in key sectors; provide targeted skilling programs for elderly workers; invest in digital infrastructure and the ageing industry; and leverage and develop elderly human capital to support inclusive and sustainable economic development.
Paper discussion/conclusion draws policy implications based on empirical finding that AI adoption mitigates negative ageing effects on GDP growth.
Robustness checks using the old-age dependency ratio as the proxy for ageing deliver consistent results.
Paper reports robustness verification: replacing the primary ageing measure with the old-age dependency ratio yields similar threshold/mitigation findings.
When AI adoption (industrial robot penetration) surpasses a critical threshold, the negative effect of ageing on GDP growth is significantly mitigated.
Threshold interaction result from panel threshold regression: AI adoption (robot penetration) as threshold variable; paper reports that beyond a critical robot-adoption threshold the negative ageing–GDP relationship is significantly weakened.
Organizational support and continuous learning are important to maximize the benefits of AI integration in startup environments.
Conclusions drawn from thematic analysis of interviews with 12 startup employees emphasizing need for organizational support and ongoing learning.
AI functions as a workforce augmentation tool that enhances human capabilities rather than replacing employees.
Reported perceptions from 12 startup employees in semi-structured interviews; thematic coding indicated view of AI as augmentation rather than replacement.
Most employees demonstrated progressive adjustment and competency improvement over time after initial adaptation.
Interview data from 12 startup employees with thematic analysis indicating progressive adjustment and competency gains over time.
AI improves employee performance by supporting more accurate decision-making and increasing work effectiveness and output quality.
Findings from semi-structured interviews of 12 startup employees, analyzed via thematic coding and frequency scoring, reporting improved decision accuracy and output quality with AI support.
AI integration contributes to competency development, particularly in digital literacy, analytical thinking, and adaptive learning.
Qualitative semi-structured interviews with 12 startup employees; thematic coding highlighted competencies (digital literacy, analytical thinking, adaptive learning).
AI significantly enhances employee productivity by accelerating task completion, reducing manual workload, and improving workflow efficiency.
Qualitative study using semi-structured interviews with 12 startup employees; data analyzed with thematic coding, frequency scoring, and visualized analysis.
Experiments highlight a reward anatomical structure that balances income, profit, efficiency, fairness, and customer retention, moving beyond income-only goals.
Experimental design / reward engineering reported in paper; claim supported by experiments (no quantitative metrics or sample size given in excerpt).
Training strength is validated by benchmarking against fixed, rule-based models and cost-plus in controlled experimentation.
Paper reports controlled experiments benchmarking ARL models against fixed/rule-based and cost-plus baselines; specific experimental design and sample sizes not provided in excerpt.
Inventory challenges are addressed by utilizing a curated dataset that has been enhanced through feature engineering, transformation, and systematic cleaning, providing reliable inputs for training.
Methodological claim about dataset curation and preprocessing used to train ARL agents; no dataset size or quantitative validation reported in excerpt.
Profitability in a dynamic marketplace is enhanced through an Adaptive Reinforcement Learning (ARL)-based pricing framework that utilizes Q-Learning and Deep Q-Networks (DQN) for real-time optimization in response to changing market conditions, competition, and inventory levels.
Paper proposes and experiments with an ARL-based pricing framework (methods include Q-Learning and DQN); validation claimed via benchmarking/controlled experimentation against baselines (details not provided in excerpt).
Dynamic pricing is crucial for maximizing revenue and maintaining competitiveness in markets with fluctuating demand, perishable goods, and diverse customer preferences.
Conceptual claim stated in paper's introduction/motivation; no empirical sample or experiment specified in the statement.
In the long term, big data promotes sustained improvements in individuals’ welfare.
Theoretical long-run growth analysis in the model showing that sustained data sharing leads to long-run welfare improvements (analytic/model-based, no empirical/sample data).
There exists an optimal level of data (big data) sharing that achieves the best balance between economic development and privacy, thereby maximizing individuals' welfare.
Analytical optimization within the theoretical macro model: model yields an interior optimum for data-sharing intensity that trades off economic gains and privacy costs (derivation/analytical result; no empirical test).
Structured intent representations (PPS) can improve alignment and usability in human–AI interaction, especially in tasks where user intent is inherently ambiguous.
Synthesis of experimental findings (rendered PPS better on goal_alignment overall, task-dependent gains concentrated in high-ambiguity business tasks) and the preliminary user survey.
A preliminary retrospective survey (N = 20) suggests a 66.1% reduction in follow-up prompts required, from 3.33 to 1.13 rounds, when using PPS.
Authors report a small retrospective survey of N = 20 respondents comparing number of follow-up prompt rounds required before vs after adopting PPS (self-reported).
We introduce goal_alignment, a user-intent-centered evaluation dimension, and find that natural-language-rendered PPS outperforms both simple prompts and raw PPS JSON on this metric.
Experimental comparison across the three prompt conditions using the goal_alignment evaluation dimension applied to the collected outputs (540 outputs across 60 tasks and 3 models), as judged by an LLM judge.
We propose a multi-agent discussion framework wherein specialized agents collaboratively process extensive product information, distributing cognitive load to alleviate single-agent attention bottlenecks and capturing critical decision factors through structured dialogue.
Method description: multi-agent discussion architecture described and implemented; claimed to distribute cognitive load and reduce single-agent attention bottlenecks (design + reported behavior).
To enhance simulation stability, we implement a mean-field mechanism designed to model the dynamic interactions between the product environment and customer populations, effectively stabilizing sampling processes within high-dimensional decision spaces.
Method description: implementation of a mean-field mechanism within the simulator; paper asserts this design stabilizes sampling in high-dimensional decision spaces (method + reported simulation behavior).
We introduce a preference learning paradigm in which LLMs are economically aligned via post-training on extensive, heterogeneous transaction records across diverse product categories.
Method description: post-training LLMs on heterogeneous transaction records across product categories to align preferences (methodological / training procedure described).
This paper introduces a Multi-Agent Large Language Model-based Economic Sandbox (MALLES) as a unified simulation framework applicable to cross-domain and cross-category scenarios.
Paper description: design and implementation of MALLES, presented as a unified framework leveraging large-scale LLM generalization for cross-domain/cross-category simulation (methodological contribution).
Human-AI systems should be designed under a cognitive sustainability constraint so that gains in hybrid performance do not come at the cost of degradation in human expertise.
Normative recommendation in the paper based on the conceptual/mathematical framework and the identified trade-off; presented as an argument rather than empirically validated policy outcome in the excerpt.
Together, these quantities provide a low-dimensional metric space for evaluating whether human-AI systems achieve genuine synergistic performance and whether such performance is cognitively sustainable for the human component over time.
Claim about the utility of the defined metrics, supported within the paper by the conceptual/mathematical framework and the proposed metric definitions (theoretical demonstration rather than reported empirical validation in the excerpt).
The paper defines a set of operational metrics: the Cognitive Amplification Index (CAI*), the Dependency Ratio (D), the Human Reliance Index (HRI), and the Human Cognitive Drift Rate (HCDR).
Explicit listing of newly proposed operational metrics in the paper; this is a descriptive claim about the paper's content (theoretical definitions), no sample size or empirical estimation provided in the excerpt.
The paper introduces a conceptual and mathematical framework to distinguish cognitive amplification (AI improves hybrid human-AI performance while preserving human expertise) from cognitive delegation (reasoning is progressively outsourced to AI).
Explicit contribution claim in the paper (description of a conceptual and mathematical framework); evidence consists of the model and formal definitions presented in the paper (no external empirical validation reported in the excerpt).
Artificial intelligence generates positive spatial spillovers for UCEE (positive effects on neighboring regions).
Spatial Durbin model reported in the abstract indicating positive spillover coefficients for artificial intelligence.
The Global Malmquist–Luenberger (GML) index and its efficiency change (EC) and technological change (TC) components stay above 1, indicating sustained efficiency gains dominated by technological progress.
GML index and decomposition results reported in the abstract based on the panel data and GML computation.
Nationally, the average UCEE index rises from about 0.3 to above 0.7 over the sample period.
Computed UCEE index results from the Super-SBM model applied to the panel of 30 provinces (2013–2022) as reported in the abstract.
SOL-ExecBench reframes GPU kernel benchmarking from beating a mutable software baseline to closing the remaining gap to hardware Speed-of-Light.
Conceptual/positioning claim made by the authors about the intended shift in benchmarking perspective enabled by SOL-ExecBench.
To support robust evaluation of agentic optimizers, we provide a sandboxed harness with GPU clock locking, L2 cache clearing, isolated subprocess execution, and static analysis-based checks against common reward-hacking strategies.
Method/tool claim in paper describing the provided evaluation harness and its engineered controls (list of features included).
We report a SOL Score that quantifies how much of the gap between a release-defined scoring baseline and the hardware SOL bound a candidate kernel closes.
Paper defines the SOL Score metric and states its interpretive meaning (fraction of gap closed between baseline and hardware SOL bound).
SOL-ExecBench measures performance against analytically derived Speed-of-Light (SOL) bounds computed by SOLAR, our pipeline for deriving hardware-grounded SOL bounds, yielding a fixed target for hardware-efficient optimization.
Methodological claim: introduction of SOLAR pipeline to compute analytic hardware-grounded SOL bounds and use of those bounds as benchmark targets, as described in the paper.
The benchmark covers forward and backward workloads across BF16, FP8, and NVFP4, including kernels whose best performance is expected to rely on Blackwell-specific capabilities.
Paper description of benchmark coverage (workload direction and data types; inclusion of kernels tied to Blackwell hardware features).
We present SOL-ExecBench, a benchmark of 235 CUDA kernel optimization problems extracted from 124 production and emerging AI models spanning language, diffusion, vision, audio, video, and hybrid architectures, targeting NVIDIA Blackwell GPUs.
Paper reports construction of the benchmark with counts: 235 CUDA kernel problems and 124 source models; descriptive dataset claim in the manuscript.
Given these findings, policymakers should favor 'strategic forbearance'—apply existing laws rather than create new regulations that could stifle innovation and diffusion of AI.
Authors' normative policy recommendation based on their interpretation of the reviewed empirical literature (risk–benefit assessment); this is a prescriptive conclusion rather than an empirical finding, so no sample size applies.
Generative AI lowers entry costs for startups, facilitating new firm entry and product development.
Cited empirical and descriptive evidence in the literature review indicating reduced development costs and faster product prototyping enabled by AI tools; the brief does not provide a pooled sample size or a single quantitative estimate.
Generative AI significantly boosts productivity in specific tasks like coding, writing, and customer service—often by 15% to 50%.
Synthesis/review of empirical literature through 2025 (multiple empirical studies of task-level impacts, including field and lab studies and observational analyses); the brief reports aggregate reported effect ranges but does not list a single pooled sample size.
The AgentDS benchmark datasets are open-sourced and available at https://huggingface.co/datasets/lainmn/AgentDS.
Paper includes link to the open-source datasets and the AgentDS website.
The strongest solutions arise from human-AI collaboration.
Analysis of competition results showing top-performing submissions employed human-AI collaborative approaches rather than AI-only baselines (results from 29 teams / 80 participants).