Evidence (14922 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
9047 claims
Filter claims →
Productivity
8066 claims
Filter claims →
Governance
7278 claims
Filter claims →
Human-AI Collaboration
6912 claims
Filter claims →
Org Design
4439 claims
Filter claims →
Innovation
4359 claims
Filter claims →
Labor Markets
3652 claims
Filter claims →
Skills & Training
3018 claims
Filter claims →
Inequality
2160 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 | 795 | 210 | 105 | 955 | 2131 |
| Governance & Regulation | 886 | 414 | 197 | 126 | 1654 |
| Organizational Efficiency | 826 | 204 | 129 | 87 | 1257 |
| Technology Adoption Rate | 681 | 259 | 128 | 110 | 1189 |
| Research Productivity | 464 | 138 | 65 | 349 | 1028 |
| Output Quality | 503 | 196 | 61 | 53 | 813 |
| Decision Quality | 351 | 180 | 84 | 51 | 673 |
| AI Safety & Ethics | 238 | 288 | 71 | 34 | 637 |
| Firm Productivity | 455 | 58 | 92 | 20 | 631 |
| Market Structure | 186 | 172 | 123 | 25 | 511 |
| Task Allocation | 222 | 70 | 76 | 34 | 407 |
| Innovation Output | 238 | 28 | 48 | 18 | 334 |
| Skill Acquisition | 177 | 62 | 62 | 17 | 318 |
| Employment Level | 107 | 57 | 108 | 13 | 287 |
| Fiscal & Macroeconomic | 135 | 72 | 44 | 26 | 284 |
| Firm Revenue | 172 | 50 | 28 | 5 | 256 |
| Consumer Welfare | 121 | 68 | 45 | 12 | 246 |
| Task Completion Time | 183 | 33 | 10 | 13 | 240 |
| Inequality Measures | 45 | 126 | 50 | 6 | 227 |
| Worker Satisfaction | 95 | 74 | 23 | 12 | 204 |
| Error Rate | 77 | 98 | 11 | 4 | 190 |
| Regulatory Compliance | 84 | 73 | 17 | 7 | 181 |
| Automation Exposure | 61 | 61 | 27 | 14 | 166 |
| Training Effectiveness | 98 | 21 | 14 | 19 | 154 |
| Wages & Compensation | 78 | 37 | 25 | 6 | 146 |
| Developer Productivity | 105 | 18 | 14 | 6 | 144 |
| Team Performance | 87 | 17 | 28 | 10 | 143 |
| Job Displacement | 12 | 83 | 23 | 1 | 119 |
| Hiring & Recruitment | 53 | 8 | 8 | 3 | 72 |
| Social Protection | 39 | 17 | 8 | 2 | 66 |
| Creative Output | 32 | 20 | 8 | 3 | 64 |
| Skill Obsolescence | 5 | 50 | 6 | 1 | 62 |
| Labor Share of Income | 17 | 20 | 17 | — | 54 |
| Worker Turnover | 15 | 15 | — | 3 | 33 |
| Industry | — | — | — | 1 | 1 |
Users treat the system as a collaborative research partner, delegating tasks such as drafting content and identifying research gaps.
Qualitative and quantitative analysis of interaction logs in the Asta dataset showing user behaviors where the system is used to draft content and identify gaps (examples and aggregated counts described in paper).
Users submit longer and more complex queries than in traditional search.
Comparative analysis of query length/complexity in the Asta Interaction Dataset (>200,000 queries) versus traditional search baselines (as reported in the paper); measurement of query length and complexity metrics across logs.
ASR-assisted transcription offers a practical pathway toward scalable, technology-supported documentation of endangered languages.
Authors' interpretive conclusion based on the corpus creation, ASR model performance (CER ~15%), and reported reductions in transcription time/cognitive load; presented as a recommendation/implication rather than a directly measured outcome.
ASR integration can substantially reduce cognitive load for transcribers.
Paper reports evaluation of ASR assistance including cognitive-load outcomes (authors claim cognitive load is reduced); details of measurement instrument, sample size, and statistical results are not given in the abstract.
ASR integration can substantially reduce transcription time.
Paper reports an evaluation of the impact of ASR assistance on the efficiency of speech transcription (comparison of ASR-assisted vs manual transcription). The abstract asserts a substantial reduction in transcription time but does not provide numeric details in the provided text.
Public Model Context Protocol (MCP) server repositories are the current predominant standard for agent tools.
Paper asserts MCP servers are the predominant standard and uses these repositories as the primary monitoring source.
Drawing on analysis of agentic investment firm operational models demonstrating 50-70% cost reductions while maintaining fiduciary standards.
Internal analysis/modeling of agentic investment firm operational models reported by the authors; paper states the 50–70% cost reduction result but provides no sample size or detailed empirical validation in the provided text.
The proposed system architectures and findings provide practical implications for future development of agentic AI systems for engineering design.
Concluding/implicational claim based on the methods and experimental findings reported in the paper (battery pack design experiments); no empirical test of 'practical implications' is provided in the excerpt.
Using machine learning applied to news streams constitutes a practical method to augment existing fiscal surveillance tools.
Paper asserts practical applicability of ML + news for surveillance; presented as recommendation/claim rather than documented large-sample trial in the provided excerpt.
Incorporating news-based signals into machine-learning models can enhance regulatory practice by improving detection of potential fiscal instabilities.
Paper claims an empirical analysis and synthesizes findings linking news-derived signals and ML methods to improved regulatory monitoring; specific datasets, evaluation metrics, and sample sizes are not provided in the excerpt.
The framework offers a replicable model for governments and institutions seeking to proactively support high-potential innovations across sectors.
Paper asserts replicability and applicability to governments/institutions based on the described methods and outputs; no deployment case studies or empirical replication evidence reported in text provided.
A data-driven, foresight-based approach to policy design significantly enhances responsiveness, precision, and resource efficiency in science and technology governance.
Paper concludes this benefit based on its integrated framework, triangulation, Delphi/AHP validation and illustrative mapping; no quantified comparative metrics or experimental evaluation reported in text provided.
Fostering digital transformation alongside workforce reskilling and innovation-ecosystem development is essential for sustainable industrial growth and strengthening Kazakhstan’s global economic position.
Policy and strategic recommendations based on the study's empirical results, case studies, and macro-level index comparisons.
Digital transformation combined with workforce retraining optimizes labor costs and enhances productivity.
Synthesis of enterprise-level case examples and aggregated regression/correlation findings at industry and national levels that link digitalization and retraining programs to labor-cost and productivity indicators.
Overall, the DRL framework enhances traffic capacity and fuel efficiency without compromising safety.
Aggregate interpretation of simulation results comparing DRL-based AV control to IDM across capacity, fuel efficiency, and safety metrics within the simulated scenarios. Specific safety metrics and sample sizes are not described in the claim text.
These findings provide an early empirical baseline and point toward competitive plurality rather than winner-take-all consolidation among engaged users.
Interpretation synthesized from survey results (multi-platform usage, indistinguishable satisfaction among top platforms, differing adoption reasons); overall sample N=388.
Switching costs between platforms are negligible (users treat these tools as interchangeable utilities rather than sticky ecosystems).
Survey responses indicating platform-switching behavior and perceived costs; inference based on reported multi-platform usage and responses about platform loyalty/switching (overall N=388).
These results establish agent scaling as a practical and effective axis for HLS optimization.
Synthesis/interpretation of empirical results (including mean 8.27× speedup and per-benchmark gains) reported in the paper.
Across benchmarks, agents consistently rediscover known hardware optimization patterns without domain-specific training.
Qualitative and empirical observations across the evaluated benchmarks (12) reporting that agents found recognized hardware optimization patterns despite no hardware-specific training.
This work demonstrates the technical feasibility of scalable, AI-augmented quality assessment for early childhood education and lays a foundation for continuous, inclusive AI-assisted evaluation enabling systemic improvement and equitable growth.
Overall results of dataset release, Interaction2Eval performance (agreement), and deployment efficiency reported in the paper; used by the authors to argue broader feasibility and potential systemic impact.
AI-assisted monitoring could shift assessment practice from annual expert audits to monthly AI-assisted monitoring with targeted human oversight.
Authors' synthesis combining dataset-scale results, Interaction2Eval performance (agreement), and deployment efficiency gains to argue feasibility of more frequent monitoring.
These findings provide quantitative foundations for AI capability-threshold governance.
Synthesis/interpretation of model results and empirical validation described in the paper (recommendation/implication).
Digital transformation enhances the relational embeddedness among cities, and this enhanced relational embeddedness facilitates improved outcomes in collaborative innovation (mediating mechanism).
Mediation analysis / network metric analysis using city-level relational embeddedness measures computed from patent collaboration networks and digital transformation indicators from A-share listed companies (2011–2021).
The work advances theory on human performance in complex negotiations and offers validated design guidance for interactive systems.
Authors' stated contributions: theoretical advancement and validated design guidance, grounded in the presented empirical results and the validated visualization tested in the N=32 experiment.
Robust arbitrage strategies remain profitable even when generalized across different domains (claim reiteration emphasizing cross-domain profitability and robustness).
Repeated/strengthened claim in the paper referencing multiple experiments and robustness checks across domains.
An arbitrageur can efficiently allocate inference budget across providers to undercut the market, creating a competitive offering with no model-development risk.
Methodological description and empirical demonstration in the paper showing arbitrageur strategies that allocate inference budget across multiple providers to create a competitive service without incurring model-development risk.
Arbitrage reduces market segmentation and facilitates market entry for smaller model providers by enabling earlier revenue capture.
Reported analysis and/or experiments suggesting arbitrage homogenizes offerings (reduces segmentation) and allows smaller providers to capture revenue earlier through arbitrage-enabled routes.
Robust arbitrage strategies that generalize across different domains remain profitable.
Reported experiments indicating that arbitrage strategies generalized beyond the primary SWE-bench domain and still yielded profit (authors state robust strategies remain profitable across domains).
Arbitrage is viable in AI model markets (we empirically demonstrate the viability of arbitrage and illustrate its economic consequences).
Empirical experiments and analyses presented in the paper (case study on SWE-bench and additional experiments on arbitrage strategies).
The paper introduces the Distributed Human Data Engine (DHDE), a socio-technical framework previously validated in biological crisis management, and adapts it for regional economic flow optimization.
Author statement describing the DHDE and asserting prior validation in biological crisis management; adaptation described in paper (methodological description).
The ACT represents the first open-source effort to consolidate data on Africa's evolving HPC landscape, aiming to encourage more transparency from local AI stakeholders and facilitate broader access for AI developers.
Authors' characterization of ACT as a novel, open-source consolidation; assertion based on literature/tools review performed by the authors and on the tool's stated goals.
This systematic framework can help predict at a detailed level where today's AI systems can and cannot be used and how future AI capabilities may change this.
Interpretive/utility claim: authors argue that the ontology plus classification results serve as rough predictive tools for AI applicability across work activities.
EnterpriseLab provides enterprises a practical path to deploying capable, privacy-preserving agents without compromising operational capability.
Conclusion drawn by the authors based on the platform design and the reported empirical results (performance parity with GPT-4o, cost reductions, benchmark robustness). The abstract offers this as a high-level takeaway rather than a quantified empirical claim.
Training humans to develop teamwork competencies, independent from task training, can enhance collaboration and performance in human-agent teams (HATs).
Overall experimental findings in KeyWe: task-independent teamwork training (<30 min) was associated with higher delegation, more strategy-based assignment, and better performance under difficulty for trained teams compared to controls.
Trained teams demonstrated resilience by achieving higher task performance when the game difficulty increased.
Performance comparison under increased difficulty in the KeyWe game between teams with trained humans and teams without training; task performance measured (score or completion metric) showed trained teams performed better under harder conditions.
This pattern suggests that AI search may make hotel discovery less exclusively controlled by commission-based intermediaries (OTAs).
Interpretation/inference from the observed higher non-OTA citation shares for experiential queries in the audited Google Gemini sample; not a direct measurement of market outcomes such as bookings or commissions.
The results contribute to literature arguing that cloud-based GenAI is a source of enterprise value creation rather than merely an experimental technology.
Paper's stated addition to the existing literature based on the combined empirical and theoretical findings.
When compared to baseline approaches, the ARL-based model's accuracy in revenue and price optimization decreased by less than 20%, indicating that it can adapt and optimize pricing techniques in intricate, cutthroat markets.
Reported experimental comparison versus baselines (fixed/rule-based and cost-plus); specific metrics, dataset size, and whether 'decrease' refers to error or accuracy are not clarified in the excerpt.
Our results substantiate the potential of large language models as a foundational pillar for high-fidelity, scalable decision simulation and latter analysis in the real economy based on foundational database.
High-level conclusion drawn from the paper's experiments and methodological contributions; generalization claim asserting LLMs' potential as foundational tools for scalable, high-fidelity decision simulation.
Experiments demonstrate that our framework achieves improved simulation stability compared to existing economic and financial LLM simulation baselines.
Empirical claim: experiments vs. baselines showing improved simulation stability (paper statement that framework improved simulation stability, without quantitative details in the excerpt).
Experiments demonstrate that our framework achieves significant improvements in purchase quantity prediction compared to existing economic and financial LLM simulation baselines.
Empirical claim: experiments comparing MALLES against existing baselines; paper reports 'significant improvements' in purchase quantity prediction (no numerical values provided in the excerpt).
Experiments demonstrate that our framework achieves significant improvements in product selection accuracy compared to existing economic and financial LLM simulation baselines.
Empirical claim: experiments comparing MALLES against existing economic and financial LLM simulation baselines; paper reports 'significant improvements' in product selection accuracy (no numerical values provided in the excerpt).
This preference-learning approach enables the models to internalize and transfer latent consumer preference patterns, thereby mitigating the data sparsity issues prevalent in individual categories.
Claim based on the paper's reported approach: cross-category post-training and transfer of latent preferences; supported by experiments (paper states mitigation of data sparsity).
Orchestrated systems of smaller, domain-adapted models can mathematically outperform frontier generalist models in most institutional deployment environments.
Formal conditions and comparative analysis derived in the paper plus referenced/claimed empirical support across several domains (frontier lab dynamics, alignment evolution, sovereign AI pressures).
Debiasing via metadata redaction and explicit instructions restores detection in all interactive cases and 94% of autonomous cases.
Intervention experiments in Study 2 where metadata redaction and explicit instructions were applied to interactive assistants (e.g., GitHub Copilot) and autonomous agents (e.g., Claude Code); reported full restoration for interactive and 94% for autonomous.
An increasing number of enterprises are using the label of artificial intelligence merely as a cosmetic embellishment in their annual reports (the phenomenon of 'AI washing' is spreading).
Framing/background claim in the paper's introduction/abstract; implied support from the semantic analysis of annual report texts across Chinese A-share firms over 2006–2024.
There are ethical imperatives of fairness and transparency in automated wealth management, and the paper proposes a roadmap toward sustainable and interpretable financial AI.
Normative analysis and proposed roadmap described in the paper; the excerpt does not provide operationalized fairness metrics, interpretability methods, or evaluation results.
In environments characterized by high-frequency data, non-linear dependencies, and stochastic market regimes, autonomous DRL agents can learn optimal sequential decision-making policies that offer a compelling alternative to static or rule-based allocation strategies.
Argument based on theoretical suitability of DRL for sequential decision problems and the paper's system-level investigation; excerpt does not report specific experimental datasets, sample sizes, benchmarks, or performance metrics.
The integration of Deep Reinforcement Learning (DRL) into portfolio management represents a significant evolution from classical Mean-Variance Optimization and modern econometric frameworks.
Conceptual comparison and synthesis presented in the paper; no empirical sample size or experimental results are provided in the excerpt to quantify the degree of improvement.
Blindfolding (anonymizing identifiers) allows verification of whether meaningful predictive signals persist (i.e., predictions reflect legitimate patterns rather than pre-trained recall of tickers).
Combined methodological-and-result claim: approach described (anonymization) plus stated objective and reported validation (negative controls and reported Sharpe under anonymization). Specific experimental protocol and quantitative results isolating the effect of anonymization are not provided in the excerpt.