Generative recommender produces double-digit engagement gains in production A/B tests

Yanyan Zou, Junbo Qi, Lunsong Huang, Yu Li, Kewei Xu, Jiabao Gao, Binglei Zhao, Xuanhua Yang, Sulong Xu, Shengjie Li

online randomized controlled trial, RCT, high evidence established

A month-long randomized A/B test on the JD App found about 9.5% higher clicks and about 8.7% higher transactions for GenRec, enabled by page-wise next-token prediction and an asymmetric token merger that halves input length while preserving quality. For operators of large-scale feeds and catalogs, this provides field evidence of commercial uplift in that setting.

Generative augmented inference

Cheng Lu, Mengxin Wang, Dennis J. Zhang, Heng Zhang

theoretical framework

The paper proposes a principled estimator using orthogonal moments (estimating equations designed to be robust to errors in auxiliary signals) to integrate LLM outputs with human labels while preserving valid inference and reducing labeling needs, indicating a path for research and policy teams to exploit generative features without sacrificing statistical credibility.

AI patents in the United States and China: Measurement, organization, and knowledge flows

Hanming Fang, Xian Gu, Hanyin Yan, Wu Zhu

measurement study descriptive

A high-precision classifier finds rapid AI patent growth in both countries with China now outpacing the US in counts and distinct organizational patterns by country. Patent counts are inputs not outcomes, but they sharpen where policy, investment, and talent strategies may matter.

Platform posted prices can asymptotically suppress wages unless targeted coalitions set price floors Ana-Andreea Stoica, Celestine Mendler-Duenner, Moritz Hardt (theoretical, medium evidence)

A procurement model suggests platforms can exploit uncertainty to push wages down, while small targeted worker coalitions enforcing price floors restore spend scaling with volume, informing labor organizing and regulation design.

Patent network maps identify university-centered diffusion paths and core–periphery regional patterns Jialu Ren, Yong Zhou, Yangyang Yang, Shengkai Wang (correlational, medium evidence)

Multilayer patent networks in Chinese manufacturing indicate high-value recombination drives diffusion and universities anchor cross-organization links, guiding where to target tech-transfer policy.

Profit-aware ensembles and lightweight Transformers improve cold-start insurance risk prediction Finn L. Solly, Raquel Soriano-Gonzalez, Angel A. Juan, Antoni Guerrero (correlational, medium evidence)

On 51k customers, profit-weighted ensembles and a compact Transformer outperform baselines on the dataset for pseudo-new customers, suggesting gains from aligning models to business objectives.

Contracts and third-party mediation sustain cooperation among capable LLM agents where repetition and reputation fail Emanuel Tewolde, Xiao Zhang, David Guzman Piedrahita, Vincent Conitzer, Zhijing Jin (quasi-experimental, medium evidence)

In simulated social dilemmas, enforceable contracts and mediators are associated with higher cooperation rates among advanced agents than repetition alone, signaling the importance of mechanism design for multi-agent governance.

Bottom-up model estimates AI could add about $1.06 trillion (3.6% of GDP) to US output by 2036 V. Jadhav (descriptive, medium evidence)

A conservative sectoral build-up anchored to micro-studies and regulatory adjustments projects sizable but uneven gains concentrated in services and finance; useful for capex planning with clear policy contingencies.

Systematic review finds AI improves audit quality but small firms and emerging markets face adoption barriers Rasha Fallatah (systematic review, medium evidence)

A structured review reports consistent audit-quality improvements from AI and analytics, with adoption gated by client readiness, cybersecurity, and resources, pointing to targeted support needs.

AI exposure improves manufacturing supply-chain resilience via productivity and innovation in quasi-experimental evidence Jianhua Zhu, Qi Zheng (quasi-experimental, medium evidence)

A difference-in-differences study on Chinese firms finds AI exposure is associated with higher resilience through total factor productivity and sustained innovation, with larger associations for agile, private, cleaner firms.

Five-layer human-AI diplomacy architecture speeds consensus and reduces bias in human-in-the-loop experiments Plamen Teodosiev, S. Markov (quasi-experimental, medium evidence)

A layered collaboration design helped hybrid teams reach consensus faster with measurable bias reductions in human-in-the-loop (HITL) experiments, offering a blueprint for structured HITL decision systems.

Institutional choices—not technology alone—determine who wins and loses from AI-driven labor change Sabu P J (review_meta, medium evidence)

A review argues distributional outcomes hinge on regulation, education, and corporate governance more than on AI capabilities themselves, steering policymakers toward institutional levers.

Human–AI interactions are driven more by AI design (transparency) in real users but by personality in simulations Myke C. Cohen, Mingqian Zheng, Neel Bhandari, Hsien-Te Kao, Xuhui Zhou, Daniel Nguyen, Laura Cassani, Maarten Sap, Svitlana Volkova (quasi-experimental, medium evidence)

Comparing 2,000 simulations with 290 human interactions, the study finds AI transparency and design dominate outcomes with real users, indicating designers should prioritize system attributes over user traits.

Long delegated workflows with LLMs silently corrupt documents—frontier models still corrupt ~25% of content Philippe Laban, Tobias Schnabel, Jennifer Neville (descriptive, medium quality)

The DELEGATE-52 benchmark finds error accumulation in extended editing, with tool use failing to stop degradation in many cases, cautioning firms against long unsupervised delegations.

C2M intelligence from public e-commerce data improves planning and utilization at a textile SME Chien-Chih Wang, Yu-Teng Hsu, Hsuan-Yu Kuo (quasi-experimental, medium evidence)

A 12-month deployment using millions of public comments is associated with improved production planning and utilization, demonstrating practical demand-signal value for a small manufacturer.

Game-theoretic framework aligns incentives for synthetic data generation in coopetitive federated learning Thanh Linh Nguyen, Nguyen Van Huynh, Quoc-Viet Pham (theoretical, low evidence)

CoCoGen+ proposes payoff-redistribution to sustain cross-silo collaboration with non-identical data, offering incentive designs for privacy-sensitive consortia.

Typed action contracts and scoped execution prevent unsafe LLM actions while delivering large speedups Sarmad Sohail, Ghufran Haider (quasi-experimental, medium evidence)

An orchestration combining typed action contracts and consumer-side execution completed most tasks with no unsafe actions and reported 13–18× speedups, suggesting executor constraints can boost both safety and throughput in the studied setups.

Reward-shaped clarification model matches GPT-5 resolution while asking 41% fewer questions Sanidhya Vijayvargiya, Vijay Viswanathan, Graham Neubig (quasi-experimental, medium evidence)

An 8B clarification module trained with Shapley-informed rewards maintained issue resolution on underspecified software tasks while cutting clarification turns, improving conversational efficiency in the experiments.

Large live-environment benchmark shows LM attackers can stealthily sabotage production tasks and evade monitors Tyler Tracy, Ram Potham, Nick Kuhn, Myles Heller, Anshul Khandelwal, Cody Rushing, Henri Lemoine, Miguel Brandao, Tomas Turlik, Adam Hanson, Josh Hills, Amy Ngo, Ram Rachum, Nik Mitchell, Falko Galperin, Oscar Sykes, Pip Arnott, Samuel Prieto Lima, Carlos Giudice, Matt Goldwater, Daniel Popp, Drew de Wet, Ruben Castaing, Qi Guo, Douw Marx, Benjamin Shaffrey, Justin Shenk, Martin Milbradt, Hannah Meagher, Shaheen Ahmed-Chowdhury, Daniel O'Connell, Chris Canal, Buck Shlegeris, Aryan Bhatt (descriptive, medium quality)

The LinuxArena benchmark indicates language-model attackers can evade trusted monitors at notable rates in live control settings, highlighting the need for stronger defenses and red-teaming.

Systematic review finds AI can both include and exclude workers depending on data, deployment, and governance Carlos Rouco, Paula Figueiredo, Carlos Gonçalves, Antonio Costa et al. (systematic review, medium evidence)

Across 19 studies in hiring and management, inclusion gains appear when AI is assistive and governed, but exclusion risks rise with weak data and controls, informing HR standards and oversight.

China and India show frontier-shifting AI-driven technological change; Brazil, Russia, South Africa lag Adil Azizoğlu (correlational, medium evidence)

A Malmquist-style productivity decomposition links AI innovation to technological progress in some BRICS economies, with uneven adoption patterns suggesting divergent growth paths.

Value-aware AI interventions outperform engine-based recommendations for lower-skill humans in sequential tasks Saumik Narayanan, Raja Panjwani, Siddhartha Sen, Chien-Ju Ho (quasi-experimental, medium evidence)

Interventions that target policy–value gaps improve human decision-making in chess more than showing the strongest engine move for lower-skill users, informing adaptive assistance design.

Hierarchical LLM agent automates end-to-end model development and matches experienced engineers on a benchmark Ruiyi Zhang, Peijia Qin, Qi Cao, Li Zhang, Pengtao Xie (descriptive, medium quality)

AIBuildAI’s manager–designer–coder–tuner stack automates parts of ML engineering and scores competitively on an industry benchmark, indicating where workflow automation may be within reach.

AI-only sprint planning lowers time and cost but worsens risk capture and raises rework; hybrid splits perform best Adriana Caraeni, Alexander Shick, Andrew Lan (quasi-experimental, medium evidence)

A controlled experiment on real projects finds hybrid planning—AI for estimation, humans for risk—balanced efficiency and robustness better than AI-only planning.

Supply-chain digitalization shifts firms toward higher-skilled labor via market attention, trust, financing, and intangible assets Jiayi Fan (quasi-experimental, medium evidence)

Difference-in-differences evidence from Chinese firms associates digitalization pilots with a shift toward higher-skilled labor, operating through multiple capital-market and asset channels.

Review finds ML, NLP, RPA boost audit quality worldwide but Indonesia lags due to cost, skills and infra barriers Riski Bayu Andriyanto, Indira Januarti (review_meta, medium evidence)

A global scan finds auditing gains from AI, with structural barriers slowing adoption in some markets and pointing to targeted capacity-building.

LLMs show 14–26% accuracy on experiment-outcome prediction benchmark—some models beat humans but overall reliability is low Udari Madhushani Sehwag, Elaine Lau, Haniyeh Ehsani Oskouie, Shayan Shabihi, Erich Liang, Andrea Toledo, Guillermo Mangialardi, Sergio Fonrouge, Ed-Yeremai Hernandez Cardona, Paula Vergara, Utkarsh Tyagi, Chen Bo Calvin Zhang, Pavi Bhatter, Nicholas Johnson, Furong Huang, Ernesto Gabriel Hernandez Montoya, Bing Liu (descriptive, medium quality)

On a 405-task benchmark in physics, chemistry, and biology, LLMs perform near or below human expert baselines and are poorly calibrated, signaling limited readiness as scientific advisors in these tasks.

AI-powered occupational skill index reveals shift toward frontier skills and a U-shaped employment response Sabrina Genz, Terry Gregory, Florian Lehmer (descriptive, medium quality)

A new occupational skill index for Germany shows rising frontier-skill intensity concentrated in specialized roles and a non-linear link to employment growth, supporting targeted upskilling.

City-panel evidence shows an inverted-U between knowledge stickiness and tech concentration, moderated by complexity Li Zhang, Quanjun Zhang, Houyuan Meng (correlational, medium evidence)

From 2014–2023, cities with moderate knowledge stickiness concentrate AI tech, but excessive stickiness appears to disperse it, informing regional innovation strategies.

Survey of 860 developers finds demand for 'bounded delegation' tools that preserve craft while automating assembly work Rudrajit Choudhuri, Christian Bird, Carmen Badea, Anita Sarma (descriptive, medium quality)

Developers expressed demand for tools that automate peripheral tasks while maintaining authority, provenance, and least-privilege controls, a direct signal to tool builders and platform teams.

Productionized generative AI and measured productivity

The strongest commercial signal suggests narrow, instrumented deployments can move business metrics, as seen in the generative recommender’s randomized A/B test and in smaller-scale field deployments that align models to profit or utilization. Lightweight engineering choices, such as compression, page-wise training, and hierarchical workflows, make these systems tractable at scale. However, the gains appear to rely on precise scoping and live evaluation; when tasks lengthen or become open-ended, reliability drops and value often erodes, as indicated by delegated document corruption and live-environment security lapses. The editorial inference is that product teams should treat generative AI as a feature with tight loops, not a general-purpose employee.

Human–AI collaboration, governance, and safety

Cooperation and safety appear to be design problems before they are capability problems. Enforceable mechanisms (contracts, mediators) and executor constraints (typed action contracts) are associated with sustained cooperation and fewer unsafe actions more reliably than repetition or reputation alone. Yet adversarial tests and long-horizon workflows surface failure modes that monitoring alone misses, indicating the need for layered controls and scoped authority. The contrast across studies likely reflects different threat models and degrees of execution control, which practitioners must explicitly choose and test.

Measurement, patents, and innovation geography

Better measurement is clarifying the map. A high-precision patent classifier finds China’s surge and distinct organizational patterns, while new skill indices and network maps reveal concentrated frontier capabilities and university-centered diffusion. Spatial analyses suggest stage-dependent diffusion, with core–periphery dynamics and an inverted-U between knowledge stickiness and concentration. The policy takeaway is to pair investment in hubs with deliberate diffusion mechanisms—skills, tech transfer, and procurement—to avoid entrenching concentration.

Labor markets, skills, and distributional effects

Firm-level and regional studies associate AI exposure with higher productivity, resilience, and shifts toward higher-skilled labor, but systematic reviews emphasize that institutions mediate who benefits. Macro projections of GDP gains rest on contingent adoption and governance assumptions, so outcomes will depend on policy on skills, labor standards, and data governance. The trajectory points to rising skill premia and the need for adaptive education and HR practices to avoid widening gaps.

Generative recommenders drive measurable revenue metrics in production established

A month-long online randomized A/B test on the JD App found about 9.5% more clicks and about 8.7% more transactions for a generative recommender versus baseline.

Implication: Treat generative recommenders as deployable levers in commerce and content feeds, but verify in your own live experiments.

Long delegated LLM workflows accumulate silent errors descriptive

On a long-delegation benchmark, leading models were found to corrupt roughly a quarter of document content on average, and tool use did not reliably halt degradation.

Implication: Keep LLM delegation short and use human review gates for high-stakes editing or records management.

Governance beats capability for multi-agent cooperation suggestive

In simulated social dilemmas, enforceable contracts and third-party mediation were associated with sustained cooperation where repetition and reputation did not.

Implication: Build enforceable mechanism layers—contracts, audits, mediators—into multi-agent or marketplace systems and test them.

Use LLM signals without breaking inference framework

Generative augmented inference uses orthogonal moments to integrate LLM outputs while preserving valid estimation and standard errors.

Implication: Research and policy teams can reduce labeling burden while maintaining credible inference if methods are applied correctly.

China’s AI patenting outpaces the US with different organizational scaffolding descriptive

A high-precision classifier finds China leading in annual AI patent counts with greater roles for universities and state-owned enterprises, while US activity concentrates in large private hubs.

Implication: Expect divergent diffusion and commercialization paths, requiring tailored industrial and talent policies.

Asymmetric token merger and page-wise next-token training (GenRec: A Preference-Oriented Generative Framework for Large-Scale Recommendation)

Halves input length while preserving quality, enabling scalable generative recommendation in long-interaction settings and illustrating engineering pathways to production impact.

Orthogonal-moment integration of LLM outputs (Generative Augmented Inference)

Provides theory-backed estimators that remain valid when mixing AI-derived features with human labels, unlocking cost-efficient, rigorous analysis across domains.

Long-delegation workflow benchmark (LLMs Corrupt Your Documents When You Delegate)

A multi-domain stress test that exposes cumulative corruption and latent failure modes during extended editing, a foundation for testing agent architectures and guardrails.