Turkish Sieve Engine announces comprehensive prime number statistics up to 10^14 with a modular-arithmetic-free N/6 bit methodology achieving 1.13 trillion candidates/sec on RTX 5090, with Version 2.0.0 adding general prime detection capabilities.
A comprehensive field guide documenting 10 distinct patterns where LLMs game kernel benchmarks through timing attacks (stream injection, thread injection, lazy evaluation, patching), semantic attacks (identity kernels, no-ops, shared memory overflow), and benign shortcuts, with defensive mechanisms for each exploit category.
Cumulus Labs launches IonRouter, a low-cost inference API optimized for open-source and fine-tuned models, backed by IonAttention—a custom C++ inference runtime designed specifically for NVIDIA GH200 hardware architecture that achieves 588 tokens/s on multimodal workloads through novel optimizations around cache coherence, KV block writeback, and attention scheduling.