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.
Search
2,125 results
181–200
Preprint reports extra tetraquark resonances with a fuller basis
An all-Jacobi calculation found new poles in strange-charm and double-charm systems and noted uncertainty in their masses and widths.
Future muon-proton collider could find a heavy T quark
Preprint: A detector-level simulation projects 5σ sensitivity up to 3.1 TeV in a T to tZ search, subject to modeled collider and detector assumptions.
UAV mapping model reports higher 3D reconstruction scores
Preprint: GeoFF3D combines georeferenced camera positions with spatial chunking and reports stronger average results on aerial mapping benchmarks.
Preprint Classifies Rota–Baxter Operators on Two Finite Groups
A 28 August 2026 arXiv preprint classifies all E3 operators and all non-trivial H3 operators, with related results on centers and bijectivity.
Iron-powder simulations produce sharply different ignition thresholds
Preprint: A computational study reports different simulated ignition patterns across model choices, coflow temperatures and burner enclosures, with thresholds differing from experiment.
Voltage clustering shrinks a year's grid operating points by 99.66%
Preprint: In a synthetic high-renewables grid model, voltage-response clustering preserved annual voltage-security behavior more accurately than injection-based and heuristic sampling.
Robot Planner Rejected One in Four Candidates in Tests
Preprint reports stronger complex-task planning on manual-guided benchmarks, but physical tests used a no-load panel simulator and the emergency stop lagged.
Preprint reports exact match between two knot-invariant constructions
The formal algebraic result connects Nichols-algebra V_n invariants with colored Links-Gould invariants, with mirror and parameter inversion tied to a right-module convention.
Preprint links solar wind Alfvénicity to radiation belt swings
An analysis of about 40 million measurements found much higher median electron flux during high-Alfvénicity wind, with the clearest seasonal coherence at solar minimum.
Algorithms offer new guarantees for multi-tier network design
Preprint presents approximation and verification methods for graphs with nested connectivity requirements, but its main randomized guarantee applies only when the number of tiers is fixed.
Vision-Language Encoders Shift Which Clues Matter During Search
Preprint: In 800 controlled scenes, frozen OpenCLIP and SigLIP often changed which visual clue was most useful as evidence accumulated.
Blockchain query system shrinks proof data, but scans still ran faster
Preprint: VeriTS uses authenticated data structures and model-based bounds to verify time-series queries across Ethereum and TRON test streams.
Optical bound states show different responses to structural changes
A preprint analysis predicts when a nearby zero remains a BIC and when it is classified as an LTM.
New analysis maps how split constraints change in optimization
Preprint: The study finds an exact sensitivity formula in one restricted case, with bounds and necessary tests for broader problems.
Sparse Node Monitoring Caught Simulated Synchronization Shifts
An arXiv preprint found that two warning signals tracked modeled transitions from a small set of monitored nodes, but the approach has not been tested on real systems.
JWST-Level Stability Mostly Falls Short in Exoplanet Telescope Model
Preprint: A tenfold reduction in modeled mirror motion reaches the target under idealized assumptions, but unmodeled vibration remains a major caveat.
Deep VISTA survey maps millions of sources in Magellanic Clouds
A preprint reports sharper, more continuous views of the Magellanic Clouds' crowded central regions.
Model predicts a teleportation-like proton spin signal at 151 MeV
A theoretical preprint models entangled three-proton scattering and finds a narrow polarization window, but reports no experimental demonstration.
Layered AI defenses stop tested jailbreaks but reject benign prompts
Preprint: A seven-layer test found no residual attacks in its run, but the stack refused 81% of benign prompts and offered no clear gain over its strongest layer.