Preprint: AI model learns to vary reasoning effort across maths problems
An arXiv preprint finds that adaptive routing reduced an AI model’s reasoning-token use on maths benchmarks while preserving near-base accuracy on MATH-500.
An arXiv preprint finds that adaptive routing reduced an AI model’s reasoning-token use on maths benchmarks while preserving near-base accuracy on MATH-500.
For years, a Python developer who needed a GPU had two realistic choices: Learn NVIDIA CUDA C++ well enough to write an extension, set up a build toolchain…
Entities: NVIDIA
An arXiv preprint reports higher Nishimori criticality in a Potts learning model, with numerical evidence and a toric-code threshold comparison.
An arXiv preprint reports model-specific collapse suppression, convergent-norm bound states and localization across a full spectrum in singular quantum equations.
A preprint study finds LFU a strong protocol-specific baseline for LLM caches, while quality-adjusted hit rates were far below raw rates on LMSYS and QQP.
A preprint reports that staged IAR training improved retrieval-free document-question and general-benchmark scores over direct fine-tuning in most main tests, with-—
A preliminary surgical AI preprint found chunked forecasting outperformed one-shot prediction for future surgical frames and instrument trajectories.
A cosmology preprint models a tiny transient dark-matter component that could enhance early halos, but data do not favor it over ΛCDM.
A preprint audit finds AI self-training gains can reflect measurement artifacts, while external distillation reaches more low-base problems than tested self-training
An arXiv preprint finds that frame choice can make warm-inflation dissipation look weaker, allowing quantum-dominated fluctuations in some high-temperature models.
A preprint tests a single-inceptor control concept for a transformational eVTOL, reporting stable simulations, early pilot feedback and important limits on the data.
Kimi Agent is a name that's come to cover a sprawling family, and untangling it matters before judging any piece of it.