Local AI Builds Collider Workflows, But Review Still Matters.
Preprint: A local-model evaluation found higher workflow pass rates after deterministic checks, but difficult requests often reached approval in altered form.
Search
2,111 results
141–160
QCD Study Finds U(1) Solution Could Leave Strong CP Unresolved
Preprint: A low-energy analysis argues that, under its assumptions, avoiding CP violation would require weakening the anomaly contribution needed to address the U(1) problem.
A4 and B4 have a special vertex, unlike D4 and F4
Preprint: Exact checks find a special vertex in A4 and B4, but not D4 or F4.
Flow Matching outperforms diffusion in sparse-view CT test
Preprint: A computational comparison found lower generation KID and stronger reconstruction metrics for Flow Matching, but included no clinical outcomes.
Preprint classifies locally finite derivations across three algebra families
The theorem-driven study separates the quantum plane, the first quantum Weyl algebra and differential extensions, with exact answers in each case.
Four of seven A stars show magnetic signatures in new study
Preprint: all four stars chosen by a 5200 Å spectral dip and TESS rotational modulation had magnetic detections, while none of the three chosen through abundance and slow rotation showed a Zeeman signature.
New yttrium lifetimes test how nuclear shapes change
Preprint: Fast-timing measurements of 16 excited states in four neutron-rich yttrium isotopes support, but do not settle, a model of changing nuclear shapes.
CMB distance data strongly favor alternative expansion models
Preprint analysis finds all three GPAge parameterizations favored over flat ΛCDM with CMB data, but the physical explanation remains provisional.
Quantum-dot charge maps generated with far fewer simulations
Preprint reports large computational savings, while agreement with experiment shifts with the assumed oxide thickness.
AI System Reports Higher Scores With First-Error Tracking
Preprint: A framework for tracing the first wrong step in an AI team reported higher scores on three benchmarks, but its limits remain untested.
Black-Hole Model Extends Beyond a Surface at Infinity
Preprint analysis finds that a theoretical black-hole spacetime takes different paths beyond infinity depending on rotation and direction.
Residual-guided neural networks show higher accuracy on 71 benchmarks
Preprint: Across UCI classification benchmarks, the method reported higher average accuracy and lower variability than three corresponding baselines.
Preprint: MoS2 films grew in four distinct forms across substrates
Under the same nominal process, crystalline surfaces were associated with different domain sizes, coverage levels and light-emission signals.
LLM-guided drone searches showed paths up to 52% shorter
Preprint: The system found three targets faster than a conventional scan in simulation, while its real-world validation was a described test with no reported trial count.
Study reports lower blood-pressure errors using buffered sample selection
A 2026 arXiv preprint reports lower errors on two public clinical datasets while leaving the underlying model and training objective unchanged.
Urban Forecasting Model Reports Lower Errors With Sparse Ride Data
Preprint testing in New York and Chicago reports that TransMod led benchmark comparisons while transferring knowledge across transport modes.
Preprint reports proof of determinant conjecture for real-node matrices
The algebraic result covers even-dimensional matrices with pairwise-distinct real nodes and every integer exponent at least one less than the dimension.
3D scan maps tissue chemistry alongside structure
Preprint: A full-field fluorescence method mapped silver, iron and barium signals in two specimens while phase-contrast tomography supplied structural context.
New mathematical framework tracks symmetries that can shift time
Preprint: A mathematical study links two reduced phase spaces and derives conserved quantities that may depend explicitly on time.
AI Argument Search Showed Better Stance Matching, With a Topic Trade-Off
Preprint: Tests of three dense embedding models found better support-versus-attack retrieval, but some fine-tuned comparisons showed weaker sensitivity to the claim topic.