2,226 results for block · 2.281s

News for “block”
17 results • 2268 ms server time
Moozonian News
kernelspace.naigap.com• Aug 12, 2026• 1 min read
Show HN: Kernelspace- interactive course on systems programming for LLM ServingHi HN,*I* built kernelspace - a free and interactive course that takes a backend engineer (like me) with java/python experience to the systems level required to work on LLM serving at scale. (Mostly an attempt for me to understand what everyone's talking about lately and also an attempt at staying up to date to current tech - career wise).It has: 1. 68 Lessons across 9 tracks, with a rust-from-zero ramp. 2. 8 rust labs - BPE tokenizer, KV block manager, radix cache, batching scheduler, a toy executor 3. 9 simulators running in browser (wasm) - roofline model, KV cache pressure, continuous batching dynamics 4. Fleet week - a simulated prod serving fleetAll progresss is local - there is no server (it's served from github pages, repo here https://github.com/praveer13/kernelspace)Feedback appreciated, and contributions welcome. I want to make this a really good course for me (and everyone else like me).* (OK, I had sol and kimi on it for most part)
Moozonian News
picklebrowser.com• Aug 11, 2026• 1 min read
Show HN: Pickle – token efficient AI agent browser with policy-gated actionsI use agents for a lot of my own work, but whenever they access my browser they end up wasting a lot of tokens reading HTML, so I tried building a browser that simplifies webpage content for them.Pickle is an agent browser and it has all the features a regular browser has (tabs, search history, bookmarks, etc.), but pages load as compact structured data instead of raw HTML to reduce token usage.Overview for those curious:It's policy-gated (blocked domains, actions that need approval like purchases) and every action is logged, so you're always able to see what your agent is doing and can take over at any time.It's also compatible with weaker/local models as it auto-routes by model strength. i.e. small local models are limited to picking actions one step at a time so they don't hallucinate actions or make up element IDs. Stronger models can plan multiple steps ahead.On the token compression side, I've been seeing around 32x less token usage.Other useful features: - shared notebook that y
Moozonian News
github.com• Aug 11, 2026• 1 min read
Show HN: Chain – A Python-like language with native inline C++Chain (Formerly Link-Lang) is a Python-syntax scripting language that was built from scratch in C++, with one feature I haven't noticed elsewhere: an 'extern "c" { }' block where you can write real C++ directly inside a script. The runtime then compiles it, catches the .so, and shares variables with the surrounding script via shared memory. An example would be: set x = 30 extern "c" { for(int i = 0; i < 5; i++) { LINK_x -= 1; } } print(x) # x is now 25, mutated by the C++ block This essentially means you get Python-like ergonomics for everyday usage, while having a direct hatch to C++ when you need raw performance or a library that isn't wrapped yet, not having to leave the file or set up a separate build step.Furthermore, Chain has dynamic typing (int/float/string/char/bool/list/dict), OOP (classes, constructor, 'this', 'new'), an ('import') module system, and try/catch. It also ships with native modules for GUI (via raylib), audio, TCP networking, and OS/shell access. This means a sc
Advertisement
Moozonian News
bing.com• Aug 10, 2026• 1 min read
Columbia is expanding Main Street lights. Here's what to knowColumbia's popular "ceiling of lights" on the 1600 block of Main Street will soon have company, as city leaders plan to illuminate more of the downtown corridor. But the future of the original canopy ...
Moozonian News
github.com• Aug 9, 2026• 1 min read
Show HN: Lumabri – What if LLMs worked like Napster?A while ago I started working on Colibrì to see if it was possible to run huge LLMs on a normal computer. The project grew far beyond what I expected, thanks in large part to the HackerNews community.That led me to a new question:What if we stopped thinking about one computer?This is the idea behind Lumabri.Instead of requiring a single machine to store and run an entire huge model, Lumabri treats a network of normal computers as a shared pool of resources.One machine might provide disk space, another compute, another a different part of the model. If a required block or expert isn’t available locally, the system can retrieve or execute it on a peer.This is particularly interesting for Mixture-of-Experts models. A model can have hundreds of billions of parameters, while only a fraction are activated for each token. Rather than moving huge expert weights over the network, Lumabri can send the small activation to a peer that already has the expert and let it execute it.The goal is for ma
Moozonian News
news.ycombinator.com• Aug 8, 2026• 1 min read
How Google makes sure your phones doesn't work without itHello HN,I've been playing around with AOSP alot and one thing i noticed is how much android depends on google even if Android itself doesnt need it. Here's everything i found out:1. Forced Integration of GPS (google play services) - This single handedly causes alot of problem to power users because without it almost nothing works.. Even if you use MicroG there's still some server side verifications that it just cannot help with.2. Sideloading Restrictions - Starting in 2026, Google requires developers to submit government IDs, pay fees, and register signing keys to distribute apps via sideloading, effectively blocking independent software distribution on certified devices.3. Ecosystem dependence. I don't need to explain this one at all i think4. Apps literally won't work. Banking apps and any app that uses play integrity will just refuse to work.What made me the most furious was Google locking down opensource apps with unverified devs. This caused an uproar in the open source communit
Moozonian News
nicholas-a-hall.github.io• Aug 7, 2026• 1 min read
Show HN: ESP32 Bowling System technical write-upHi everyone! I'm the ESP32 Bowling guy from a few weeks ago, back with a technical write-up and a repo link.Since my last post, I've received -tons- of support and encouragement for the project. I'm absolutely blown away. I didn't think there would be much interest!This last week, I spent a ton of time working on the prototype. It can trigger my pinsetter machines and read ball state, there's a basic state machine for the game and a React UI for the bowlers. I've also built the prototype uart bridge which connects an ESP32 gateway node to a raspberry pi lane compute module, which acts as a bridge between the ESPNOW mesh and lane-external services.The last major blocker is object detection on an ESP32-CAM module. I could go with a webcam wired to the Pi, but that feels like cheating. I really want the Pi running the state machine, websocket and REST API endpoints only. One of my key constraints for this project is that each node serves one purpose only, and it performs that purpose well
Moozonian News
github.com• Aug 7, 2026• 1 min read
Show HN: Oans – fast deduplication for btrfs and XFSHi all, I've been working for a while now on a fork of duperemove, which finds identical files/blocks and uses the filesystem's feature to deduplicate them.My goal was to make it more convenient and faster, and more or less by accident it is now about 10 times faster in the deduplication phase.The main trick is this: before sending data through the kernel's FIDEDUPERANGE ioctl, I simply read the data to prime the page cache. For some reason this results in an enormous speedup.My fork has several other differences to the original, e.g the hashfile stores configuration, it's easier to automate, plenty of bug fixes (I backported the most severe ones), and a pretty UI.I'd be interested in any feedback!
Advertisement
Moozonian News
news.ycombinator.com• Aug 6, 2026• 1 min read
Discord Support System is protecting the HackersToday my mail got hacked, where they had access for 30 min before I did the security changes. It was an old mail which I do not use often.I was supposed to get a recovery link, but did not. I tried to register with my mail for Discord Support, but could not.The hackers exploited the weaknesses of their hopeless bureaucratic of their system, by blocking everything from Discord before going into my Discord to change the mail.Now that I finally have understood it, Discord won't suddenly answer me and help me. I am following their bullshit protocol while the hackers have the access of my account right now.I think even fucking Temu has a better support system than Discord.
Moozonian News
news.ycombinator.com• Aug 6, 2026• 1 min read
Stripe Lands in MinecraftWhat is the deal with the stripe lands? 8 million blocks until the stripe lands insint that far. FIX IT MOJANG. Anyone know how to move in the stripe lands?/goto?url=CAESZQHuR6pNElxUnz8Hoo_YIURlI-beQ54EtDFVRKNwoXxxhWD5f-ZUJuXi24z87NwdYPek2kR48kTouSe9sJlYYvf0XvFnkmQ0C0Ewf3-CUdshdwsUrIMYuugtANLgjcdNN6ztvqgx
Moozonian News
theopenlake.com• Aug 5, 2026• 1 min read
Show HN: ExANS – Lossless KV cache compression at 622 GB/s on H100Hi HN,We are the developers of OpenLake, an open source storage engine for KV cache offloading to remote disk and memory.Once we offloaded to local disk, we realized the bottleneck is the PCIe or NIC bandwidth. We wondered whether on GPU lossless compression is viable for fast reads and lower TTFT.BF16 is usually very hard to compress, (high entropy of sign/mantissa). What surprised us is that real world KV blocks are very different. The exponent byte has a very low entropy and barely populated. Instead of compressing the whole tensor, we compress only the exponent stream on the GPU.We see the following results: (H100, production KV snapshot):- 1.51× lossless compression - 622 GB/s median GPU decodeDecompression is ~10× faster than a 400 Gb/s NIC bandwidth delivering data losslessly without quality change.We've are open sourcing this as: ExANS which will be available through our vLLM and SGLang connectors on OpenLake v0.8 version. No changes are required in the inference engine.I'm cur
Moozonian News
github.com• Aug 4, 2026• 1 min read
Show HN: PleasantriesPre-hook scripts for AI coding CLIs that block pleasantry-only prompts like "hi", "hello", "ok", "thank you", etc. Send a real task, not a greeting.
Moozonian News
mixar.app• Aug 4, 2026• 1 min read
Show HN: Blender for AI AgentsCreating 3D is hard. LLMs seem to be getting better at tool use and spatial understanding. While MCPs have proved to be a good way to use these tools- the current methods have these challenges:- Access to scene graph and core C modules of Blender- Lack of parallelism, only way is to run blender headless- Lack of deterministic and fast verification layer- Inference stack- only way to use inference is to hook another MCPWe're building Mixar, think Cursor for 3D. One access point to all generative inference, an agent to build scenes/blockouts, do boring stuff like UVs and export standard formats(.glb/gltf/obj/gbx/usd etc.)Try it: https://mixar.app - first week is free. Would love to get your feedback.
Moozonian News
news.ycombinator.com• Aug 4, 2026• 1 min read
Why remote roles are region specific and not 100% remote?I always wondered, because I'm based out of India and trying to secure a good remote role, why most companies offering remote positions are region‑specific—like remote roles only in the US or the UK. Why don’t they open the doors to the entire world when they’re looking for someone to work for them remotely?Another pro of it is people from different parts of the world will require less pay comparatively. Time difference? That's a reall issue — but anything besides South East Asia or Ocenia countries could pretty much overlap.I mean, I'm based out of IST, I can overlap EST all 8 hours and PST close to 5 hours.If the blocker is some regulations, why not hire on contractual basis then?Just some things I wanted to always learn.