The open‑source llama.cpp library released version b11713 on Oct 17, 2026, with speed‑up for low‑bit matrix‑vector ops on CUDA GPUs and new binary builds.
NoteBy OossaPublished by Oossa: 1 min read
The ggml‑org team published llama.cpp version b11713 on Oct 17, 2026. The update improves CUDA performance for low‑bit types by using a faster MMVQ algorithm. It also adds a small fix for HIP (AMD) GPUs on Q3_K and Q6_K quantizations. Pre‑built binaries for macOS, iOS, and several Ubuntu targets (CPU, Vulkan, CUDA) are now available for download.
Developers can run quantized LLMs on NVIDIA GPUs more quickly, lowering inference costs for local AI applications.
Models
Claude traced the hidden seed for NIST’s ECDSA curves to a simple C random generator, ending a long‑standing cryptography question.
Models
Meta updated its terms on Oct. 17, 2026 to delete language that many saw as permission to scrape news sites for AI training.
Models
Researchers found that a single honesty cue dramatically improves AI models’ willingness to flag mistakes in their own reports.
Oossa · Newsletter
The week in AI, explained
Every Monday: the stories worth knowing, in plain language. Free, no spam.
