Approximate membership queries in streams often need recent-window semantics rather than membership over all items ever seen. This paper studies guarded epoch Bloom filters, a sliding-window alternative to counting and stable Bloom filters. The structure partitions a fixed bit budget into rotating epochs, inserts only into the current epoch, clears whole segments at epoch boundaries, and keeps one additional guard epoch. This guard yields a dete... Read more ›
Rust’s SIMD abstractions were not as safe as I’d like. Until now. Read more ›
The tl;dr has spoilers, so I've put it .Also feel free to skip any of the chapters because the post turned out to be very long. I think you can read almost any of them on their own, and you can skip to On practicality if you want the big picture.The title is stolen from Scott Alexander's post . Basically, it will be my opinion on applied rationality and its successes and failures. The title speaks for itself: I think there are few successes. A few weeks before Scott's post, Eliezer wrote wher... Read more ›
Proposal to accomplish full decentralization of Tangled. Read more ›
Fixing stale site files on Bunny, working with Zensical for my secondary site, website readability issues. Driving an EV for the first time, and F1 updates. Read more ›
Fig (Ficus carica L.) is an early-domesticated fruit crop of high cultural and economic value. However, high-quality genome resources for evolutionary and genetic diversity studies remain limited. Here we report a high-quality, chromosome-level genome assembly for fig cv. ‘Green Peel’ generated with Oxford Nanopore long reads and Hi-C data, and polished with Illumina reads. The final assembly is 306.83 Mb and comprises 13 pseudo-chromosomes, with a contig N50 of 5.77 Mb. Read mapping supporte... Read more ›
CPUs are cheap. Stop pretending they aren’t. A case for decentralized architecture. 2026-06-13 Read more ›
Mixture-of-Experts (MoE) based large language models (LLMs), such as Qwen and DeepSeek, have recently emerged as an effective approach to improving model capacity without proportionally increasing computational cost. By replacing the conventional feed-forward network in dense LLMs with a set of experts and activating only a subset of them for each input token, MoE models significantly increase the total number of parameters while keeping the per... Read more ›
Des chercheurs de l'université de Californie à San Diego, épaulés par Google, viennent de prouver un truc contre-intuitif : un Pixel mis au rebut il y a trois ans tient encore tête à un serveur professionnel sur certains calculs, au point qu'on peut en assembler un vrai data center au lieu de le foutre à la poubelle. L'idée a été posée sur le . Une fois l'appareil ouvert, les chercheurs retirent tout ce qui ne sert plus, l'écran, la batterie au lithium, les caméras et la coque, jusqu'à ne gar... Read more ›
At long last, cyberpunks, adventurers -- Neofeud 2 is now released! Read more ›
La Légion d’honneur ne suffit pas à faire de Sophia Aram, l'humoriste du Crif, une spécialiste. Mais ça fait joli à la télé. Read more ›
Riding to and from my mom & sister’s house takes me to the Renton Transit Center twice daily, and I’ve been intrigued by a vegetarian restaurant across the street, named simply Blossom. So what the heck, a few days ago I stopped there for lunch. The place is kinda big, with probably more than a […] Read more ›
Rust png crate, also known as image-png, implements PNG encoding and decoding in safe Rust. It is compliant with the third edition of the PNG specification, including APNG support. Read more ›
Explore the mechanics of gradient synchronization in PyTorch distributed training, focusing on MPI primitives like All-Reduce and core techniques like pipeline parallelism, tensor parallelism, and Read more ›
In a pretty big surprise for the AI world, China’s Zhipu AI’s latest model, GLM-5.2, just grabbed the top spot in one of the most respected creative coding benchmarks out there. On June 19, 2026, Design Arena announced on X that GLM-5.2 has taken the #1 position in their single-round HTML web design leaderboard (non-agent […] The post appeared first on <a rel="nofollow" href=" Read more ›
How Biology Helps Shape Choice in Behavioral Economics The field of behavioral economics is facing several challenges and opportunities this century; the discipline is trying to find its own identity and position within science and applications. This includes debates around behavioral economics’ most fundamental idea—the causes of human decision-making. Human biology has a more important […] Read more ›
A behavioral classification system for machine-to-machine APIs. Read more ›
A simple, powerful coding workflow for AI assistants: plan-driven, risk-first, git-versioned plans you can pause and resume. Open source under Apache 2.0. Read more ›
Stop Merely Pointing at Ideas • Zettelkasten Method Most people take notes by merely pointing at ideas rather than engaging with them, which produces storage instead of knowledge. The cure is a genuine struggle to capture the complete essence of a single idea, written the way an essayist tries to find their way from A to B. The struggle is not a flaw in the process; it is the process. Read more ›
In large-model distributed training, especially large language model workloads, gradient All-Reduce increasingly stresses the memory and communication path. This paper asks whether a Compute Express Link (CXL) memory controller can aggregate low-bit gradient signals as gradient cache lines pass through it, while preserving a 32-bit floating-point (FP32) path for workloads, layers, or phases that should not use low-bit approximation. We present N... Read more ›