Visualizing Attention, a Transformer's Heart [video]
Demystifying attention, the key mechanism inside transformers and LLMs.
The latest blog posts to take off on Hacker News. 49,208 posts.
Demystifying attention, the key mechanism inside transformers and LLMs.
A few years ago, my cat gave me my most memorable middle of the night software engineering incident. I was working at a startup, and we didn’t have a formal on-call rotation yet. That was a deliberate decision, since being on-call is painful, and the team was good about just collectively keeping an eye out for urgent…
A few years ago, I have blogged about my approach to sandboxing less-trusted applications that I have to or want to run on my main machine. The approach has changed since then, so it is tim...
Unix like systems with X11 or Wayland All Unix like systems with either X11 or Wayland are well supported since ever. Linux with X11 and now Wayland is for a long time the primary system on that Kate work happens. Over the years it was, like most of the KDE applications, ported to various BSD variants. Be it some…
I have recently released an open source tool called sotrace . It maps out the .so dependencies for a binary or library, and presents th...
Why every manager should talk to their employees about stock options
Why an Intel Mac’s up to 40 Gb/s from Thunderbolt 3 is less than an Apple silicon Mac’s up to 40 Gb/s from USB4, and how you can benefit from it.
Miguel de Icaza's Blog
Defending Mastodon from Donald Trump's social media
With the recent release of OpenBSD 7.5, I decided to run through my personal OpenBSD “installer” for laptop/desktop devices. The project is built off of the dwm tiling window manager and only installs a few basic packages. The last time I updated it was with the release of 7.3, so it’s been due for an minor rework.
wingolog: article: tree-shaking, the horticulturally misguided algorithm
While I was looking the other way, Python got advanced static types! Here’s matrix multiplication, describing the input shapes and its output shape:
Swift for C++ Practitioners, Part 1: Intro & Value Types
As should be expected, a number of individuals have come forward with responses to the initial blog post. While I do not expect everyone to take even a passing interest in the rabbit-hole of the si…
I built a magical meme search engine using siglip/CLIP and vector encoding images. It was a fun way to learn about this powerful technology. I'm sharing the …
less is more
[custom_views_and_hn_comments hn_id=”40015185″] My startup Truss (gettruss.io) released a few LLM-heavy features in the last six months, and the narrative around LLMs that I read on Hacker News is now starting to diverge from my reality, so I thought I’d share some of the more “surprising” lessons after churning…
On my journey to learn CUDA, I decided to tackle the One Billion Row Challenge with it.
Issues with Kagi's AI focus, finances and leadership
By the way, you can listen to me read this post aloud on my Patreon, which starts at $1/month! My 🌶️ spicy take 🌶️ is that I don’t like seeing tech talks live. I get very little out of 95% of…
Explaining why Renovate is going to be my solution to keeping up-to-date with dependencies and it's not likely to change any time soon.
For over thirty years, a dead rabbit has hung inside Seven Sisters tube station, and thousands of people walk past it every day without noticing.
This is my journey that started as an experiment to see how my Threads feed would look like on Mastodon and ended with me finding experiences that went above and beyond my expectations.
The U.S. Cybersecurity and Infrastructure Security Agency (CISA) said today it is investigating a breach at business intelligence company Sisense , whose products are designed to allow companies to view the status of multiple third-party online services in a single dashboard. CISA urged all Sisense customers to reset…
DESCENT is a classic 3-D first-person shooter for DOS that was released in 1995, developed by Parallax Software and published by Interplay Productions. DESCENT is a contemporary of the other very popular FPS of the time, DOOM by id Software, released just over a year before. However, it is not based on DOOM; the…