Blogging in 2025: Screaming into the Void

Posted at December 04, 2025 Before social media became what it is today I used to blog a lot. And I wasn’t the only one, many people did. There was this idea of a decentralized and open web: everyone had their own little space on the web (with a selfhosted blog, or a platform like […]

How Brussels writes so many laws

how-brussels-writes-so-many-laws

The central puzzle of the EU is its extraordinary productivity. Grand coalitions, like the government recently formed in Germany, typically produce paralysis. The EU’s governing coalition is even grander, spanning the center-right EPP, the Socialists, the Liberals, and often the Greens, yet between 2019 and 2024, the EU passed around 13,000 acts, about seven per […]

NeurIPS best paper awards 2025

The Best Paper Award Committee members were nominated by the Program Chairs and the Database and Benchmark track chairs, who selected leading researchers across machine learning topics. These nominations were approved by the General Chairs and Next Generation and Accessibility Chairs. The best paper award committees were tasked with selecting a handful of highly impactful […]

The Ofcom Files, Part 4: Ofcom Rides Again

the-ofcom-files,-part-4:-ofcom-rides-again

This is a continuation of the Ofcom Files, a series of First Amendment-protected public disclosures designed to inform the American and British public about correspondence that the UK’s censorship agency, Ofcom, should prefer to keep secret. See Part 1, Part 2, and Part 3. We heard from Ofcom again today. The agency writes: The full […]

Trick users and bypass warnings – Modern SVG Clickjacking attacks

Clickjacking is a classic attack that consists of covering up an iframe of some other website in an attempt to trick the user into unintentionally interacting with it. It works great if you need to trick someone into pressing a button or two, but for anything more complicated it’s kind of unrealistic. I’ve discovered a […]

Coca Cola has an executive dedicated to McDonald’s

Roberto Mercade is president of The McDonald’s Division (TMD) of The Coca‑Cola Company. He leads a global organization that is responsible for the company’s key relationship with McDonald’s in more than 100 markets. Mercade has been with Coca‑Cola since 1992, when he began his career as a production services manager in Puerto Rico. He went […]

State of AI: An Empirical 100T Token Study with OpenRouter

state-of-ai:-an-empirical-100t-token-study-with-openrouter

This empirical study offers a data-driven perspective on how LLMs are actually being used, highlighting several themes that nuance the conventional wisdom about AI deployment: 1. A Multi-Model Ecosystem. Our analysis shows that no single model dominates all usage. Instead, we observe a rich multi-model ecosystem with both closed and open models capturing significant shares. […]

Thoughts on Go vs. Rust vs. Zig

Aug 09, 2025 I realized recently that rather than using “the right tool for the job” I’ve been using the tool at the job and that’s mostly determined the programming languages I know. So over the last couple months I’ve put a lot of time into experimenting with languages I don’t get to use at […]

Django 6

django-6

Content Security Policy support¶ Built-in support for the Content Security Policy (CSP) standard is now available, making it easier to protect web applications against content injection attacks such as cross-site scripting (XSS). CSP allows declaring trusted sources of content by giving browsers strict rules about which scripts, styles, images, or other resources can be loaded. […]

CUDA-L2: Surpassing cuBLAS Performance for Matrix Multiplication Through RL

cuda-l2:-surpassing-cublas-performance-for-matrix-multiplication-through-rl

🥳 Introduction CUDA-L2 is a system that combines large language models (LLMs) and reinforcement learning (RL) to automatically optimize Half-precision General Matrix Multiply (HGEMM) CUDA kernels. CUDA-L2 systematically outperforms major matmul baselines to date, from the widely-used torch.matmul to state-of-the-art NVIDIA closed-source libraries (cuBLAS, cuBLASLt-heuristic, cuBLASLt-AutoTuning). Paper Speedup of CUDA-L2 over torch.matmul, cuBLAS, cuBLASLt-heuristic, and […]