./daily — ← back to field notes

The link sheet

One-line pointers to today's local-AI, model, hardware, and healthcare-AI news. No commentary — just the headline, the link, and where it came from.

Tuesday, 2026-10-27·last updated 2026-10-27 19:06 EDT·9/9 feeds ok

[ hardware ]11

• Microsoft Maia 300 Shown at OCP 2026 — STH
• b11936 — llama.cpp
• b11935 — llama.cpp
• b11934 — llama.cpp
• b11933 — llama.cpp
• b11932 — llama.cpp
• b11930 — llama.cpp
• b11929 — llama.cpp
• b11927 — llama.cpp
• Tensor Split Over PCIe 3.0 x1 — Good Results — r/LL
• ASRock Creator Radeon AI Pro R9700 Review: Workstation Inference, AMD Style — STH

[ new models ]10

• OpenAI's major results over the last 3 months came from scaling test-time compute with a… — r/LL
• Statistically Lossless Qwen3.8 27B GGUFs — r/LL
• Build specialized multimodal models with LFM3 — HF
• Can we trust LLMs to judge whether a research idea is novel? — HF
• b11937 — llama.cpp
• 4 coding agents on one RTX 3090: ~290 tok/s total from a custom CUDA engine for Qwen3.8… — r/LL
• Ling-3.1-flash-1019 weights are out, and they released the base model too — r/LL
• b11928 — llama.cpp
• inclusionAI/Ling-3.1-flash-1019 · Hugging Face — r/LL
• We put DeepSeek in charge of four Qwen3.5-9B models to reverse engineer binaries using s… — r/LL

[ local ai ]12

• New audio model: LFM3-Audio-2B — r/LL
• The reason why 5090's are $10k now... — r/LL
• LiquidAI/LFM3-Omni-2B-GGUF — r/LL
• Real-time video captioning with LFM3-VL-2B running locally in-browser on WebGPU — r/LL
• OpenBMB released a 1.3B OCR model and a separate 2B model to check its work — r/LL
• LiquidAI/LFM3-1.7B · Hugging Face — r/LL
• I am a King, but local 😂 — r/LL
• A 1.3B OCR model just hit 93.68 on real-world document parsing and there's now an AI jud… — r/LL
• 8x R9700 System Upgrade — r/LL
• Spark-X2.6-1.7B, almost the same perf as 4b — r/LL
• How to Build a Decision Model Like Jev — r/LL
• Comparing ISTA-DASLab's DeepSeek-V4.1-Flash-1017-GSQ-RCO-GGUF 2.25bpw against the offici… — r/LL

[ benchmarks & papers ]2

• Guidelines for evaluating structure-based methods for protein–ligand binding prediction — NatMI
• Efficient equivariant framework for universal charge density prediction — NatMI

[ tools ]2

• Survival LLM: AI assistant and emergency library that runs on your own computer — HN
• SenseNova 6.8 Code2Video: an open-sourced model that writes and revises code to make vid… — r/LL

[ healthcare ai ]12

• STAT+: Missouri AG warns Lilly to stop requiring data from hospitals seeking 340B drug d… — STAT
• STAT+: After two deaths in autoimmune cell therapy trial, Artiva is shutting down — STAT
• STAT+: FDA offers new details on quicker approval pathway for personalized gene therapie… — STAT
• STAT+: RFK Jr. goes for broke in Iowa — STAT
• STAT+: Nonprofit resurrects shelved gene therapy after startup shuts down — STAT
• STAT+: Pharmalittle: We’re reading about a Novo setback in court, a House probe into big… — STAT
• A possible treatment for brain-eating amoebas, found in the sea — STAT
• Opinion: Emergency pediatric care is now in permanent crisis mode — STAT
• Opinion: Cancer vaccines are finally here. How do we get them to patients? — STAT
• STAT+: Pennsylvania Dutch country, tired of waiting for biopharma, builds drug company o… — STAT
• STAT+: Roche gets a lung cancer win with its KRAS drug divarasib — STAT
• STAT+: Nonprofit Bon Secours Mercy Health strikes $5.7 billion deal for surgery center c… — STAT