═ RobotNurse

./daily — ← back to field notes

Daily links

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.

Thursday, 2026-09-10·last updated 2026-09-10 04:56 UTC·8/9 feeds ok (HN)

[ hardware ]12

Qwen3.8-Flash-Next on 2x3090 + DDR4, part 4: 2.2-2.5x faster prefill by kicking the expe… — r/LL
Running qwen 3.8 27B iq3 xxs on RTX 3060. — r/LL
Don't let FOMO win if you're interested in local llm from a hobby/learning aspect — r/LL
Server rebuild to custom loop. 2x RTX Titans 24gb, 1x 22gb 2080ti | T: 70GB VRAM. — r/LL
b10883 — llama.cpp
Arm Neoverse CSS N4 Launched for Next-Gen CPUs and DPUs — STH
b10881 — llama.cpp
SOTA ImageGen Locally NVIDIA Cosmos3(64B) INT4 quants CUDA/MLX — r/LL
b10878 — llama.cpp
b10877 — llama.cpp
b10876 — llama.cpp
b10875 — llama.cpp

[ new models ]9

I made a way to migrate between embedding models without re-embedding your entire corpus — r/LL
b10884 — llama.cpp
Mention if a "new model" is a finetune — r/LL
IBM releases SOTA Granite Time Series PatchTST-FM-r2 model with commercial-friendly lice… — HF
Qwen3.8-27B-Uncensored-Genesis-V1-GGUF — r/LL
1-bit 27B in the browser: 25–30 tok/s on a 6 GB RTX 3060 Laptop (WebGPU, no install) — r/LL
DeepSeek-V4-Flash-Vision-Exp (285B MoE) on 10-12x RTX 3090 — spec decoding, vision — r/LL
Is anyone working on conversation compaction? — r/LL
Is there a dummies guide for setting up qwen 27B with dflash2 and n-gram? — r/LL

[ local ai ]12

So relevant — r/LL
Local LLM / Qwen 3.8 win — r/LL
Apple A20 Pro debuts with 7-core GPU, 32-core Neural Engine and 50% more memory bandwidt… — r/LL
Surveillance plagiarism by OpenAI — r/LL
Would you consider 5t/s usable for a local model? — r/LL
I trained an audio model that can generate infinite one-shots for music production and t… — r/LL
Best Open source TTS right now for narration? — r/LL
Desert Ant Labs: On-device intelligence for every product — r/LL
Why the hell is LM Studio making LM Studio so difficult to download? — r/LL
Now this is a serious local machine — r/LL
Deepseek Has Soft Retired Deepseek V4 Pro — r/LL
new Nex model — r/LL

[ benchmarks & papers ]2

GLM 5.3 Flash Q4 @ 60tps / 550tps on M3 Ultra — r/LL
Causal evidence that language models use confidence to drive behaviour — NatMI

[ tools ]0

Nothing notable today

[ healthcare ai ]12

all days → · 47 links · generated from 9 sources