Based: Simple linear attention language models
BASED is a linear attention architecture that decodes without a KV-cache, delivering a 24x throughput improvement over Transformers with FlashAttention-2.
The latest blog posts to take off on Hacker News. 80,900 posts.
BASED is a linear attention architecture that decodes without a KV-cache, delivering a 24x throughput improvement over Transformers with FlashAttention-2.
In this blog I explore how I use an incident management mindset to manage a complex medical condition. I hope you enjoy it!
This blog post explores an alternative formulation of Rust’s type system that eschews lifetimes in favor of places . The TL;DR is that instead of having 'a represent a lifetime in the code, it can represent a set of loans , like shared(a.b.c) or mut(x) . If this sounds familiar, it should, it’s the basis for polonius…
Updated October 2025 to reflect the growth in binaries from 12 to <50.
How do git worktrees help you work on more than one branch at the same time, and how does that differ from virtual branches in GitButler?
This 1.5 release is unlike any of the previous ones. It brings many new features that can improve quality and the general audio experience. That is achieved through machine learning. Although Opus has included machine learning — and even deep learning — before (e.g. for speech/music detection ), this is the first time…
We’re building an open source alternative to AWS. For IPv4 assignment and firewall rules, we use Linux’s Netfilter / Nftables. This subsystem provides a powerful way to handle packets addressed to the host. We recently came across flowtables - a network acceleration feature in the Linux kernel that works like a…
A (directed) graph is a set of nodes, connected by arrows (edges). The nodes and edges may contain data. Here are some graphs: All graphs made with graphviz (source) Graphs are ubiquitous in software engineering: Package dependencies form directed graphs, as do module imports. The internet is a graph of links between…
Feeling overwhelmed? Don't know where to start? Do literally anything.
The Claude 3 model family—Haiku, Sonnet, and Opus—sets new industry benchmarks across a wide range of cognitive tasks.
Apple today announced the new MacBook Air with the powerful M3 chip, taking its power-efficient performance and portability to a new level.
Some time later, I came across Iavor�s "jokes" page, including a funny bit called � The Evolution of a Programmer � in which the traditional imperative "Hello, world" program is developed through several variations, from simple beginnings to a ridiculously complex extreme. A moment�s thought turned up the factorial…
Sleek, supersonic, and short-lived. Here’s how Singapore Airlines’ Concorde service to London fell victim to politics, nationalistic jealousies, and soaring costs.
On February 14, 2024, the Court of Appeal of Paris issued an order stating that Orange, a major French telecom provider, had infringed the copyight of Entr’Ouvert’s Lasso software and violate…
Open Office Sucks
What happens if you take motion blur past its logical extreme? Here are some fun observations and ideas I encountered while trying to answer this question, with an attempt to apply the results in a procedural animation.
Among computing pioneers, Dame Stephanie "Steve" Shirley is one of the most celebrated for building a $3 billion empire in the 1960s with an all-female staff.
At work, one of the things I do pretty often is write print generators in HTML to recreate and replace forms that the company has traditionally done handwritten on paper or in Excel. This allows the company to move into new web-based tools where the form is autofilled by URL parameters from our database, while getting…
The secrets that will make you much better at listening to a degree that your relationships are significantly more successful, peaceful, gratifying, intimate, and trusting, have to do with what you…
The International Atomic Energy Agency (IAEA) is developing new and rapid methods to rapidly screen and authenticate the origin of foods like extra virgin olive oil.
Mike Bostock, Announcing: Observable Framework: Today we’re launching Observable 2.0 with a bold new vision: an open-source static site generator for building fast, beautiful data apps, dashboards, and reports. Our …
AI, science, education, & ethics
Marketing, solicitations, advertising, political contact, direct mailers, email campaigns, cookies, ads — it's all opt out. It puts the burden on you to opt out of each communication each message, each outreach from a loyalty program. It can feel like weeding a garden, but that garden is your attention. The weeds keep…
Improving the performance of a PC cooling system by writing a custom control loop in Python, and monitoring the results with Grafana.
Among a certain tech set, Kubernetes has earned a reputation as an unnecessarily complicated time-sink that startups should avoid. Using Kubernetes with a small team is seen as a sign of over-engineering.
I love hot dogs and I'm from Chicago so we knew this was coming
East Asian businesses often go out drinking.
This week brings sad news for our Duxbury Systems family. Our colleague, David Holladay, passed away in a hospital on Thursday, February 15th, following an accident at home that morning. David was a cancer survivor and valiantly battled multiple medical problems over his last five years.