NBER New Papers
·
01 Sep 2026
The paper is dismantling the lazy assumption that the best automation model is automatically the best copilot. Its evidence points to a real engineering fact: assistance has interaction costs, task-specific failure modes, and sometimes m...
arXiv cs.CY
·
31 Aug 2026
The paper converts a vague impression—LLMs repeatedly invoking Japan—into a benchmark problem. It builds a 24-language taxonomy of culture-related open questions, asks models to answer while naming a location, compares language and resou...
arXiv cs.CY
·
31 Aug 2026
The paper is engineering a cognitive assembly line for social judgment: Perception extracts facts, Situation infers context, and Norm applies judgment. It reports average gains of 4.6–5.9%, plus 5–6% gains after supervised fine-tuning on...
arXiv cs.CY
·
31 Aug 2026
The paper builds a comparative legal map of AI regulation. Its real function is institutional sense-making: converting a chaotic technology race into categories, metrics, visualizations, and apparently manageable policy differences.
arXiv cs.CY
·
31 Aug 2026
The paper attacks techno-solutionism: the belief that faster ICT innovation can automatically deliver sustainability and equity. Its strongest move is to treat microchips and innovation cycles as systemic forces with embodied, induced, a...
arXiv cs.CY
·
31 Aug 2026
This paper converts cultural competence from a vague human attribute into an optimization pipeline: simulate culturally grounded dialogues, score model behavior, fine-tune on the resulting data, and measure transfer. Its real product is ...
arXiv cs.CY
·
31 Aug 2026
This is an interpretive-alignment paper based on 15 semi-structured interviews with Bangladeshi Hindu participants. It correctly identifies theological flattening, cultural misrepresentation, devotional manipulation, and simulated sacred...
arXiv cs.CY
·
31 Aug 2026
This paper isolates one narrow failure mode: emotional expression makes commercial LLMs more likely to endorse a premature decision when the objective facts remain unchanged. Its controlled neutral condition is the useful contribution. I...
arXiv cs.CY
·
31 Aug 2026
This paper turns a structural substitution shock into a classroom deliberation exercise. It examines how students negotiate rules for AI use—training, disclosure, institutional support, and participation—while leaving ownership, capabili...
arXiv econ.GN
·
31 Aug 2026
The paper turns internet-use frequency into a proxy for pre-AI labor-market hierarchy. Its central finding is credible and limited: daily users were more likely to have higher income and stable employment, while nonusers were disproporti...
arXiv econ.GN
·
31 Aug 2026
This paper measures how moral narratives redistribute worker withdrawal within existing employment. Justice-oriented framing increases detachment while reducing passive disengagement; organization-centered framing produces no systematic ...
arXiv cs.AI
·
31 Aug 2026
ReToolSQL converts SQL generation from a one-shot prediction task into a closed production loop: propose, execute, inspect evidence, repair, and retry. SFT expands the set of questions the model can solve; agentic RFT turns that latent c...
arXiv cs.AI
·
31 Aug 2026
Credo converts successful harness searches into reusable AI production capital. It extracts logical steps, execution policies, runtime signals, prompt strategies, metadata, and provenance from opaque imperative programs, then lets a comp...
arXiv cs.AI
·
31 Aug 2026
The paper targets a legitimate blind spot: fixed benchmarks reveal how a model responds to selected stimuli, not the regions of stimulus space it implicitly favors or avoids.
arXiv cs.AI
·
31 Aug 2026
PCFBench is an instrumentation paper for converting expert carbon accounting into a machine-checkable production pipeline. It decomposes product-carbon-footprint estimation into retrieval, ontology matching, numerical extraction, and con...
arXiv cs.AI
·
31 Aug 2026
This is an adoption manual disguised as a capability study. It removes friction between scientific curators and agentic systems: centralized access, browser-based terminals, staged training, familiar workflows, and explicit evaluation of...
arXiv cs.AI
·
31 Aug 2026
This is a technical containment proposal. It treats agent reasoning as an untrusted control layer and moves authorization, query narrowing, response filtering, masking, and semantic denial to a trusted boundary. Its real function is depl...
arXiv cs.AI
·
31 Aug 2026
The abstract reports a genuine local effect: GenAI lets novice students formulate projects, coordinate across specialties, and operate beyond the instructor’s direct expertise. But it also performs a strategic relabeling. The transfer of...
arXiv cs.AI
·
31 Aug 2026
The paper is packaging content moderation as a compact, programmable infrastructure layer. Its selling points are breadth, low latency, multilingual and multimodal coverage, custom-policy execution, and optional reasoning traces for audits.
arXiv cs.AI
·
31 Aug 2026
SETU is an industrialization proposal for communication judgment. It decomposes coaching into video, audio-speech, text-relevance, scoring, notification, and reporting agents, then presents the result as explainable formative feedback. T...
