Proposed framework sets stricter rules for AI in evidence reviews
Preprint: ARISMA treats AI as an inspected, benchmarked, logged and reversible assistant, with humans retaining control over consequential review decisions.
Search
2,230 results
641–660
New Monte Carlo analysis finds finite-accuracy gains, not high-accuracy gains
Preprint: An HFHRMC analysis finds a finite-accuracy benefit from position diffusion, while its leading high-accuracy order matches optimized ULD.
Simulation finds strongest galaxy bias in node outskirts
Preprint: Simulated central galaxies had the strongest reported bias in node outskirts, while mean filament bias fell from short to long filaments.
Preprint argues gas equations carry their own arrow of time
An analytic construction reports a larger warmer-to-cooler molecular flux and a net momentum flow in the same direction, offering a response to Loschmidt’s paradox.
Benchmark Finds Faster LLM-Generated Code for Some Database Queries
An arXiv preprint reports a 2.11× speedup in a 22-query TPC-H test, with results varying by model and query.
Preprint proves sharp area bound for simple polygons
A machine-checked theorem says one area gap is at least twice another, while a related optimum remains unresolved.
Exact norm sampling hinges on Hadamard matrices
Preprint: The unrestricted minimum is N when a Hadamard matrix of order N exists and N+1 otherwise; in two stated cases, strictly positive weights require more nodes.
Math Preprint Finds Large Families of Littlewood Counterexamples
A mathematical construction finds counterexample pairs with a badly approximable first coordinate and a full-dimensional set of qualifying first coordinates.
Preprint reports faster LLM decoding with high-bandwidth flash
A trace-driven simulation reported higher per-GPU decode throughput, lower modeled energy and fewer GPU packages than baseline designs, but no hardware test was reported.
Students Report More Weekly AI Use Than Faculty and Administrative Staff
Preprint: A survey at one large university specializing in teacher education found that most respondents reported some AI use, but weekly use was much more common among students.
A Lightweight Adapter Shows Uneven Results in Five Pruned Language Models
Preprint: In descriptive tests, SHIFT-LLM had higher average accuracy than pruned-only models in most reported model-criterion comparisons, but results varied by model and layer-selection strategy.
Edge Skillguard reports zero wrongful actuation at a software boundary
Preprint: A workshop concept paper reports 0/60 wrongful actuation and 60/60 benign success at an adapter boundary; device adapters were not triggered.
Neutrino model maps JUNO damping to an ALP coupling scale
Preprint analysis finds an energy-independent dephasing pattern, but the mapped 90% limit is inherited from a reference model.
Ideal Force Kicks May Save Work in Rapid Inertial Models
A preprint reports a short-time advantage for kicked protocols, with exceptions in some quantum dynamics and at longer durations.
Five AI Models Form a Statistical Tie on Mental-Health Benchmark
Preprint: A clinician-reviewed set of 610 HealthBench conversations found no statistically clear winner among five top-scoring models, while two models showed reproducible empty refusals.
Dutch study reports lower right-wing voting with more immigrant exposure
Preprint analysis finds the association was more negative in dense areas, while low-density regions and poorer high-exposure neighbourhoods showed different patterns.
Raman study finds pressure-linked changes in Na2Co2TeO6
Preprint: Spectral features and polarization patterns changed across pressure ranges, with the strongest proposed transition markers above 13.8 GPa.
Preprint reports a possible η′ mass shift inside carbon nuclei
The low-momentum excess persisted across several fits, but the quoted reduction depends on a density model.
Parker Solar Probe filter lifts signal-to-noise for Jupiter radio data
A preprint reports cleaner archival spectra, independent polarization agreement and a near-predicted timing match, but says its screen is not a definitive detector.
Preprint finds sharp three-step cover for 2-colored complete bipartite graphs
Preprint: Every 2-colored complete bipartite graph can be covered by two monochromatic subgraphs, each with diameter at most three.