Ken Hudson
@shareastronomy
Followers
825
Following
16K
Media
92
Statuses
13K
Learning AI / ML - sometimes the hard way. Huge fan of Apple MLX. Amateur astronomer focused on spectroscopy.
Oregon City, OR. US
Joined April 2010
Clustering NVIDIA DGX Spark + M3 Ultra Mac Studio for 4x faster LLM inference. DGX Spark: 128GB @ 273GB/s, 100 TFLOPS (fp16), $3,999 M3 Ultra: 256GB @ 819GB/s, 26 TFLOPS (fp16), $5,599 The DGX Spark has 3x less memory bandwidth than the M3 Ultra but 4x more FLOPS. By running
80
353
2K
NVIDIA engineer just contributed an update that improves the generation performance of llama.cpp by up to 40% on DGX Spark. All numbers in the post above have been updated https://t.co/kSQ7Ikr2Ug
github.com
The PR #16308 sets as device property by default integrated = false to disable host buffers. While this change is needed, the additional memory copies introduce multiple cudaStreamSynchronize calls...
1
5
42
Fun fact: Zed is actually _more_ native than Microsoft's own @code, which runs on Google's Chromium 💁♀️ Learn more about the key Windows features:
zed.dev
From the Zed Blog: Zed for Windows is finally here. Download it today.
7
23
479
It's finally here. Zed is available on Windows. Built from scratch, and rendering at 1 million pixels/millisecond. Download it today 👇
175
364
4K
Sometimes we forget that NVIDIA wins because it's a software company. DGX Spark is a reminder of that. It's a CUDA dev machine that's beautiful enough and small enough to be on my desk and with enough memory to fit a truckload of params. It's not the fastest or best at anything,
✨ We were honored to deliver some of the very first NVIDIA DGX Sparks to AI Pioneer @ylecun and AI Researcher, @soumithchintala, from @Meta and @NYUniversity in NYC. “Every PhD student in AI should have one of these,” said Yann. We couldn’t agree more. We are anticipating
34
68
1K
NVIDIA sent me preview hardware of their new DGX Spark 128GB ARM64 4TB "AI supercomputer" - it's a very neat little device, here are my notes so far https://t.co/mEPHW0rgxY
simonwillison.net
NVIDIA sent me a preview unit of their new DGX Spark desktop “AI supercomputer”. I’ve never had hardware to review before! You can consider this my first ever sponsored post …
25
38
417
Here are some long-awaited performance numbers for DGX Spark using llama.cpp https://t.co/1MFTJCcnt7
github.com
Overview This document summarizes the performance of llama.cpp for various models on the new NVIDIA DGX Spark. Benchmarks include: Prefill (pp) and generation (tg) at various context depths (d) Bat...
13
40
275
Write that blog post now... without leaving the terminal 💯 🌀 blogr — A static site generator for creating and managing blogs. ⚡ Write in Markdown, preview instantly and deploy to GitHub Pages in seconds. 🎨 Includes built-in themes. 🦀 Written in Rust & built with
2
13
94
🤖 💻 LangCode CLI A powerful dev tool that unifies AI models for smart coding. Built with LangChain, it offers automated tasks and safe code changes through intelligent model routing and diff previews. Check it out! 🚀 https://t.co/TMBNWchi9T
9
40
256
Vibe coding hack from @ProgramWithAi: Instead of generating complex code, have AI output simple YAML that describes what you want. Then build a CLI to execute it. Result: deterministic, auditable AI behavior without reading a single line of generated code. Link in 🧵
5
30
468
Introducing RND1, the most powerful base diffusion language model (DLM) to date. RND1 (Radical Numerics Diffusion) is an experimental DLM with 30B params (3B active) with a sparse MoE architecture. We are making it open source, releasing weights, training details, and code to
102
259
1K
Qualcomm is acquiring Arduino, and they're launching a new SBC (Uno Q) combining a Dragonwing SoC with an Arduino microcontroller in the Uno form factor. What implications does this have for makers and education? Quick video covering the news:
33
99
713
Really well written, well thought out post. Highly recommended.
Here's my longer argument in favor of "vibe engineering" as the other end of the scale from vibe coding:
0
0
0
I know this first hand after teaching the course for a couple of semesters and setting up a teaching lab with 16 M4 Max Studios. I’m having students learn not only the basics of ML/AI but also how to setup their own Studio machines or small desktop cluster with MLX / MLX-LM.
@ronaldmannak @tim_cook Exactly, students love it because it is hands-on and direct instead of working on machines hidden behind the vaults of HPC data centers; they are working on machines and the MLX framework that they know they can setup on their own easily when the time comes. It is empowering.
2
5
12
It's funny how people are never happy. 2 years ago we were happy AI could fill in a function in for us. Now people get upset if a CLI tool can't one shot what we're trying to build. Or maybe it's a good thing we're always working to make things better.
1
1
5
Comet Lemmon is brightening and moving into morning northern skies. Besides Comet SWAN25B and Comet ATLAS, Comet C/2025 A6 (Lemmon) is now the third comet currently visible with binoculars and on long camera exposures. Comet Lemmon was discovered early this year and is still
7
186
878
Happy to host the research papers and datasets on https://t.co/EjWqnCIaXl & https://t.co/G2mDae0uv9 if someone from the community has the time the lead this project
huggingface.co
This is bad news. This is very bad news. The US government has stopped the funding of PubMed, the most comprehensive database of biomedical literature. Why would you do this? This database it literally a matter of life and death.
9
17
158
We’ve focused on improving Claude’s skills in defensive cybersecurity. The results of this are visible in Claude Sonnet 4.5, which is comparable or superior to Opus 4.1 in cybersecurity tasks—yet both faster and cheaper. Read more:
anthropic.com
How we've improved Claude's cyber defense skills
13
26
282
We’re at an inflection point in AI’s impact on cybersecurity. Claude now outperforms human teams in some cybersecurity competitions, and helps teams discover and fix code vulnerabilities. At the same time, attackers are using AI to expand their operations.
98
216
2K