@culurciello.bsky.social
Intelligence architect, father and husband, also: AI, computing, deep learning, machine learning, hardware, SW, HW
Excited about making microchips:
www.zerotoasiccourse.com/post/excited...
Frustrated about making microchips:
www.youtube.com/watch?v=pxQC... - see from time 2.50s
www.zerotoasiccourse.com/post/excited...
Frustrated about making microchips:
www.youtube.com/watch?v=pxQC... - see from time 2.50s
Excited by silicon!
I was recently asked what excites me most about the exploding open-source silicon ecosystem. Honestly? It depends who’s asking.
For one person I might talk about Basilisk - the PULP team’s seriously i...
www.zerotoasiccourse.com
November 25, 2025 at 3:45 PM
Excited about making microchips:
www.zerotoasiccourse.com/post/excited...
Frustrated about making microchips:
www.youtube.com/watch?v=pxQC... - see from time 2.50s
www.zerotoasiccourse.com/post/excited...
Frustrated about making microchips:
www.youtube.com/watch?v=pxQC... - see from time 2.50s
Computer programming jobs are going to disappear very quickly in the next year.
To add to the video below, I add most of the computer coding jobs on the list. Now every good engineer is a 10x engineer with AI coding tools. Profound change in the industry!
x.com/i/status/199...
To add to the video below, I add most of the computer coding jobs on the list. Now every good engineer is a 10x engineer with AI coding tools. Profound change in the industry!
x.com/i/status/199...
60 Minutes on X: "AI could wipe out half of all entry-level white-collar jobs and spike unemployment to 10% to 20% in the next one to five years, predicts Anthropic CEO Dario Amodei. https://t.co/K53oCApIeF https://t.co/6BgAlQ30pM" / X
AI could wipe out half of all entry-level white-collar jobs and spike unemployment to 10% to 20% in the next one to five years, predicts Anthropic CEO Dario Amodei. https://t.co/K53oCApIeF https://t.co/6BgAlQ30pM
x.com
November 17, 2025 at 3:53 PM
Computer programming jobs are going to disappear very quickly in the next year.
To add to the video below, I add most of the computer coding jobs on the list. Now every good engineer is a 10x engineer with AI coding tools. Profound change in the industry!
x.com/i/status/199...
To add to the video below, I add most of the computer coding jobs on the list. Now every good engineer is a 10x engineer with AI coding tools. Profound change in the industry!
x.com/i/status/199...
Very interesting article on how to compress Large Language Models Llm tokens: arxiv.org/abs/2510.27688
Continuous Autoregressive Language Models
The efficiency of large language models (LLMs) is fundamentally limited by their sequential, token-by-token generation process. We argue that overcoming this bottleneck requires a new design axis for ...
arxiv.org
November 6, 2025 at 8:40 PM
Very interesting article on how to compress Large Language Models Llm tokens: arxiv.org/abs/2510.27688
Want to simulate your data center?
euge-blog.github.io/2025/10/31/d...
euge-blog.github.io/2025/10/31/d...
October 31, 2025 at 2:41 PM
Want to simulate your data center?
euge-blog.github.io/2025/10/31/d...
euge-blog.github.io/2025/10/31/d...
I could not find a data center / rack / cluster AI supercomputer simulator that I liked, so I wrote one:
github.com/culurciello/...
The interesting thing to notice here is that if you want to serve inference at scale on open models like Qwen3-235B then they are mostly compute bound.
github.com/culurciello/...
The interesting thing to notice here is that if you want to serve inference at scale on open models like Qwen3-235B then they are mostly compute bound.
GitHub - culurciello/dc-simulator: a data center cluster and rack simulator for advaced LLM models
a data center cluster and rack simulator for advaced LLM models - culurciello/dc-simulator
github.com
October 29, 2025 at 5:05 PM
I could not find a data center / rack / cluster AI supercomputer simulator that I liked, so I wrote one:
github.com/culurciello/...
The interesting thing to notice here is that if you want to serve inference at scale on open models like Qwen3-235B then they are mostly compute bound.
github.com/culurciello/...
The interesting thing to notice here is that if you want to serve inference at scale on open models like Qwen3-235B then they are mostly compute bound.
We live in a time that is often described as “digital-first.” And what about the physical?
euge-blog.github.io/2025/10/28/p...
euge-blog.github.io/2025/10/28/p...
Physical vs digital
Physical vs digital
euge-blog.github.io
October 28, 2025 at 7:10 PM
We live in a time that is often described as “digital-first.” And what about the physical?
euge-blog.github.io/2025/10/28/p...
euge-blog.github.io/2025/10/28/p...
Claude code sessions are getting smaller and smaller. They will hook you and then bleed you
October 17, 2025 at 4:10 PM
Claude code sessions are getting smaller and smaller. They will hook you and then bleed you
Engineered boys / girls bands or AI music? In recent years music is not grown from talent, it is grown in a greenhouse, or a cell-culture lab. But catchy music is designed to be catchy and will catch you! Talent? www.funktasy.com/pop-music/bo...
Boy Bands and Girl Groups - The Shift from West to East | Funktasy
Boy bands and girl groups have not always had the best reputation, however, it is impossible to deny that they have impacted the music scene.
www.funktasy.com
August 23, 2025 at 9:03 PM
Engineered boys / girls bands or AI music? In recent years music is not grown from talent, it is grown in a greenhouse, or a cell-culture lab. But catchy music is designed to be catchy and will catch you! Talent? www.funktasy.com/pop-music/bo...
Memory for AI? See: semianalysis.com/2025/08/12/s...
Scaling the Memory Wall: The Rise and Roadmap of HBM
The first portion of this report will explain HBM, the manufacturing process, dynamics between vendors, KVCache offload, disaggregated prefill decode, and wide / high-rank EP. The rest of the repor…
semianalysis.com
August 14, 2025 at 2:48 PM
Memory for AI? See: semianalysis.com/2025/08/12/s...
Wow what a difference from my super-BAD experience with Azure, I was up and running in 10 mins on www.vultr.com with a docker-based AI cloud inference image! Really recommend these guys!
SSD VPS Servers, Cloud Servers and Cloud Hosting
Vultr Global Cloud Hosting - Brilliantly Fast SSD VPS Cloud Servers. 100% KVM Virtualization
www.vultr.com
August 13, 2025 at 3:06 PM
Wow what a difference from my super-BAD experience with Azure, I was up and running in 10 mins on www.vultr.com with a docker-based AI cloud inference image! Really recommend these guys!
MS Azure is soooo bad.. even their simplest tutorial DO NTO WORK learn.microsoft.com/en-us/azure/...
Quickstart - Deploy Docker container to container instance - Azure CLI - Azure Container Instances
In this quickstart, you use the Azure CLI to quickly deploy a containerized web app that runs in an isolated Azure container instance
learn.microsoft.com
August 12, 2025 at 5:26 PM
MS Azure is soooo bad.. even their simplest tutorial DO NTO WORK learn.microsoft.com/en-us/azure/...
What do you recommend as OS / App AI tool to store all your data, including interactions with online LLMs (ChatGPT, Claude, Gemini, Ollama, etc). Maybe something that can also fill forms online and on PDF for you based on your data?
August 12, 2025 at 2:54 PM
What do you recommend as OS / App AI tool to store all your data, including interactions with online LLMs (ChatGPT, Claude, Gemini, Ollama, etc). Maybe something that can also fill forms online and on PDF for you based on your data?
Boy setting up machines on Azure or AWS today is a no-code trap. Seemingly simpler to do that in CLI, but at the end they do not work so you do 2x the work. First learn the no-code flow, fail, then have to go back to CLI and do by hand... lol progress
August 8, 2025 at 5:59 PM
Boy setting up machines on Azure or AWS today is a no-code trap. Seemingly simpler to do that in CLI, but at the end they do not work so you do 2x the work. First learn the no-code flow, fail, then have to go back to CLI and do by hand... lol progress
Training a large language model #LLM on thousands of GPUs is a daunting task. Learn some of the internals here: docs.pytorch.org/tutorials/in...
Large Scale Transformer model training with Tensor Parallel (TP) — PyTorch Tutorials 2.8.0+cu128 documentation
docs.pytorch.org
August 8, 2025 at 5:43 PM
Training a large language model #LLM on thousands of GPUs is a daunting task. Learn some of the internals here: docs.pytorch.org/tutorials/in...
AI acceleration of economies of scale - I think this article below misses an important point. Once AI will be able to do 50%+ of human jobs, society will split into poor that will never again find jobs, and rich, who can use AI and produce more.
July 28, 2025 at 12:30 PM
AI acceleration of economies of scale - I think this article below misses an important point. Once AI will be able to do 50%+ of human jobs, society will split into poor that will never again find jobs, and rich, who can use AI and produce more.
Not sure why companies continue to have CS-like coding interview with their candidates in AI... when everyone today does not need to know ANYTHING about coding to be really productive. I guess this is the proxy for "IQ test"... may be ok for kids out of schools... lol
July 25, 2025 at 5:57 PM
Not sure why companies continue to have CS-like coding interview with their candidates in AI... when everyone today does not need to know ANYTHING about coding to be really productive. I guess this is the proxy for "IQ test"... may be ok for kids out of schools... lol
Well this is just great, you can do coding Agent at home maybe soon: github.com/QwenLM/Qwen3... question is: what hardware / GPU can run this?
GitHub - QwenLM/Qwen3-Coder: Qwen3-Coder is the code version of Qwen3, the large language model series developed by Qwen team, Alibaba Cloud.
Qwen3-Coder is the code version of Qwen3, the large language model series developed by Qwen team, Alibaba Cloud. - QwenLM/Qwen3-Coder
github.com
July 24, 2025 at 7:43 PM
Well this is just great, you can do coding Agent at home maybe soon: github.com/QwenLM/Qwen3... question is: what hardware / GPU can run this?
Do you recommend working with UpWork or other freelance service sites on large projects and for experienced senior engineers? What is your experience with the platform and what tips / tricks / insights do you have for beginners?
July 23, 2025 at 1:04 PM
Do you recommend working with UpWork or other freelance service sites on large projects and for experienced senior engineers? What is your experience with the platform and what tips / tricks / insights do you have for beginners?
Reigniting the debate: stocks or real-estate / rentals investments: www.reddit.com/r/RealEstate... RE ~ 2x STOCKS. RE wins.
From the RealEstate community on Reddit
Explore this post and more from the RealEstate community
www.reddit.com
July 21, 2025 at 6:50 PM
Reigniting the debate: stocks or real-estate / rentals investments: www.reddit.com/r/RealEstate... RE ~ 2x STOCKS. RE wins.
AI for 3d models: this is really good - app.scenario.com you can generate a 3d model with distinct parts! A game-changer for 3d assets and. game development
app.scenario.com
July 21, 2025 at 6:33 PM
AI for 3d models: this is really good - app.scenario.com you can generate a 3d model with distinct parts! A game-changer for 3d assets and. game development
Dear happy Internet, I am glad you are having fun reading about Astronomer and the concert... but please keep your eyes on target: www.huffpost.com/entry/donald... one day you will have a daughter and protecting her will be your life goal
Why Donald Trump’s Jeffrey Epstein Cover-Up Matters
The conspiracism that propels MAGA has turned on its host.
www.huffpost.com
July 20, 2025 at 3:44 PM
Dear happy Internet, I am glad you are having fun reading about Astronomer and the concert... but please keep your eyes on target: www.huffpost.com/entry/donald... one day you will have a daughter and protecting her will be your life goal
Now that Windsurf is basically going down the drain, what do you guys recommend to use to code? Claude?
July 18, 2025 at 7:05 PM
Now that Windsurf is basically going down the drain, what do you guys recommend to use to code? Claude?
US citizen are missing out on cheap everyday needs because their gov is dead on keeping zombie alive www.bloomberg.com/news/feature...
Made-in-China Chevys for $17,000 Are Winning Fans in Mexico
Small cars subject to high tariffs aren’t an option for cash-strapped American buyers.
www.bloomberg.com
July 18, 2025 at 7:01 PM
US citizen are missing out on cheap everyday needs because their gov is dead on keeping zombie alive www.bloomberg.com/news/feature...
Windsurf merger and acquisition scandal: www.youtube.com/watch?v=wSud... - geek out on the craziness of AI startups
Windsurf x Google x Cognition: Full Breakdown: Who Made Money, Who Did Not
YouTube video by 20VC with Harry Stebbings
www.youtube.com
July 18, 2025 at 1:09 PM
Windsurf merger and acquisition scandal: www.youtube.com/watch?v=wSud... - geek out on the craziness of AI startups
You guys have to see this. What a crazy blunder! youtu.be/6EkmhXgBWIM?... The worst of mergers and acquisitions, the worst for talent in the company... and the crazy egotistical move of founders leaving everyone behind. Such a DRAMA Windsurf acquihire and acquisition.
Pump.Fun raises $600 million in 12 minutes & Meta's 200 million hire
YouTube video by NFA Podcast with Nish & Frackers
youtu.be
July 18, 2025 at 12:41 PM
You guys have to see this. What a crazy blunder! youtu.be/6EkmhXgBWIM?... The worst of mergers and acquisitions, the worst for talent in the company... and the crazy egotistical move of founders leaving everyone behind. Such a DRAMA Windsurf acquihire and acquisition.