sing - Moozonian Search

con-dog.github.io• Jul 27, 2026• 1 min read

Show HN: Watch 14-Byte AI "brains" attempt to solve a 2D maze (Its hard)Hey HackerNews,I built this project over the last few weeks as a palette cleanser from a failed game launch.I wanted to learn a bit about AI/Neural-Networks and naively thought I could build a tiny maze-solving AI in a weekend with a 100% solve rate.Well - I couldn't, but I got pretty close. 14 Bytes total model size, and a 96.5% solve rate on unseen mazes. Trained across 46 phases experimenting with different ideas to improve the model (better performance, smaller size).Its quite fun to watch the model attempt to solve the maze, when they fail its usually due to getting stuck in a loop. The models have no access to coordinates, map-data, or external memory scratches - they must navigate using only immediate local neighbourhood observations.There is a model dropdown and you can see how the model has progressed over each phase, constantly getting smaller and increasing its solve rate. Total trained models number in the thousands - I just expose the winning models from each phase.Overall

bing.com• Jul 27, 2026• 1 min read

ServiceNow's Revenue Growth Is Accelerating: Is the Stock a Buy Now?The stock is down 37% this year, but management is raising its full-year guidance.

bing.com• Jul 27, 2026• 1 min read

Carly Simon Reveals Parkinson's Disease DiagnosisCarly Simon has been diagnosed with Parkinson’s disease. The 83-year-old singer-songwriter, best known for her 1972 hit “You’re So Vain,” revealed the news on Monday in a statement (via People). “So ...

github.com• Jul 27, 2026• 1 min read

Show HN: Injectable time and timers with custom Date types for TypeScriptI find TypeScript js/node environment is missing a Time-Provider (just like *.*NET TimeProvider for instance ships with its framework).So for this purpose I built a small library (@time-provider) tailored for js/node environment.Obviously the goal is to both avoid using global monkey-patching with fake-timers or re-writing your clock interface... and still use your date library (native or custom).I propose four clock strategies: system (real time), fixed manual (advance time explicitly), and sequential (replay of a sequence of timestamps).I also benchmarked slightly better performances than jest / sinon.The project is still young, so the API may still receive refinements as real usage appears.I would be happy to hear feedback on this ! Take care.You can check it out on GitHub: https://github.com/jaenyf/time-providerJaenyf

news.ycombinator.com• Jul 27, 2026• 1 min read

Ask HN: 1950's Chip industry undermined unions how is AI not doing the same?In the 1950s the Chip industry undermined minorities, immigrants, women and killed unions.In the 1960-70s it targeted black civil rights groups.Along the way, white men were pushed into management, while anything outside of business was split into smaller tasked jobs. This in return gave companies leverage to cut salaries.I don't care what altman says chatgpt did in a blackbox. I judge the world by what I see. Reality is, chat, claude, etc etc are about as useful as smart as the stupidest, slowest engineer on the team.On the other hand, introducing them has made the job appear more "streamline" more production line like. Work increased, and salaries flattened.Now that companies are not getting returns off of AI, they are cutting AI spending, but make no mistake. Salaries will not recover.What am I missing? How am I wrong?Would love opinions on this

bing.com• Jul 27, 2026• 1 min read

Vector ThrustAll the Latest Game Footage and Images from Vector Thrust Vector Thrust is a fast-paced single and mulltiplayer combat flight game, using a unique cel shaded art style. With an extensive variety of ...

github.com• Jul 27, 2026• 1 min read

Show HN: RelativeDB – OSS query engine for relational foundation modelsStructured data could be the next big frontier for foundation models.RelativeDB is an open source query engine built around the Relational Transformer architecture, developed by Stanford’s Tabular and Relational (STAR) Project. These models can adapt to new databases and prediction tasks using relatively little labeled context.With some of my experiments, the results have been remarkable. On some tasks, RelativeDB can outperform XGBoost with only a small amount of relational context and no manual feature engineering.So RelativeDB turns relational foundation models into something you can query (like this customer churn query):``` PREDICT NOT EXISTS(orders.*) OVER (90 DAYS FOLLOWING) FROM customers ```The engine finds the relevant context across the related tables and returns a prediction. The goal is to make questions like churn, demand forecasting, fraud detection, and recommendations feel more like database queries than bespoke ML projects.Longer term, I think the optimizer should be

hiitmeup.fit• Jul 27, 2026• 1 min read

Show HN: Simple HIIT timer for efficient workoutsHi HN community,As a busy techie with kids and other family obligations and my recent health diagnosis, I wanted an efficient way to integrate workouts in my daily routine and my research indicated that HIIT/Tabata style routines fit that requirement much better than a dedicated long running/planned workout routine.I made a simple PWA to help me with that endeavor that I have found useful and hoping some of you may like it as well. No sign ups needed but you can sign in if you'd like to save custom presets and history/progress. You may also enjoy some of the fitness guides/blog posts that I am put together to help someone get started more easily. Built using javascript, vite and firebase.Feedback is most welcome, both via the app/email or comments here.

portzero.net• Jul 27, 2026• 1 min read

Show HN: Port Zero – How I learned to stop worrying and love PORT=0Hi HN,Recently I wasted several hours wrangling my dev environment only to find out that the browser frontend was talking to the wrong version of the backend. This got me thinking--why on earth are we still using simple numbers to describe which process to connect to? Why not use names instead? I thought of all the times a program wouldn't start because of port conflicts. The more I thought about it, the crazier it seemed.Modern operating systems already offer no-conflict ports: if you make your TCP server listen on port 0, the OS assigns you a random available port. But that only solves half the problem.I built the other half: PortZero. It's a GPLv3 program that watches for processes (and docker containers!) with a special PZ_TUNNEL environment variable: PZ_TUNNEL=myapp-{branch}.portzero.local:80 npm run dev When it sees such a process or container, PortZero does this:1. Create a virtual NIC (if it hasn't already)2. Create a new virtual IP address3. Create a DNS record that substitute

bing.com• Jul 26, 2026• 1 min read

Best PEMF mats in 2026: Which one is right for you?PEMF (pulsed electromagnetic field) mats have become increasingly popular among people looking to support relaxation, recovery, and everyday wellness at home. However, while brands often make ...

ggtranslator.com• Jul 26, 2026• 1 min read

Show HN: GG Translator – Turn gaming shit talk into friendly phrasesFlaming teammates usually results in them cooperating even less. But sometimes you just need to let the devil out of you. Instead of venting on your teammates and pissing them off, vent into GG Translator and it will convert your toxic message into a friendly and constructive one to share with your team. https://ggtranslator.comMy background is in building web apps with Elixir and Phoenix. I made GG Translator in early 2024 to see if I could build a desktop app in a language I had never used before (Rust) with the help of AI. It was a great exercise and I found out that I could. I used Tauri instead of Electron because it has a smaller footprint. I built the app but never released it because I wasn't ready to support it.When GPT 5.6 came out earlier this month I pointed it at the old codebase and tasked it with rewriting that entire app (still in Rust using Tauri), modernizing, refactoring, and looking for any improvements that it could find.---The app is a sneak peek of how an AI prof

github.com• Jul 26, 2026• 1 min read

Show HN: Codex fixed my 2010 MacBook Pro driver troubles with kernel patchesI installed Gentoo on my mid-2010 Macbook Pro about a year ago as both a fun learning project and to have a practical secondary laptop for terminal access, light dev, using utilities that would be more cumbersome on my Windows laptop, etc.It landed me with a usable KDE environment with acceleration, but sleep mode resulted in an unrecoverable system without a reboot and the graphics driver had oddities like garbled screens during login and occasionally crashing the system if I opened the wifi settings. At the time I tried switching to older proprietary nvidia drivers and some other possible solutions for sleep issues, including asking an older GPT model at some point to investigate the system, but they were dead ends - it seemed to simply be an issue of unsupported old hardware, and I don't have the time to learn driver dev.But when 5.6 Sol came out and I tried it on Ultra, I was struck by how capable it seemed, so I figured I had nothing to lose throwing it at fixing my system by any

twitter.com• Jul 26, 2026• 1 min read

Show HN: Maginary.ai gets seedance2 and GPT-images2 supportpresenting maginary -- an image/video generator with a midjourney-like prompt syntax, using 40+ underlying models and acts like an abstractized openrouter for multimedia with an amazing ux, can try it right now with no ccthis launch presents a major update: integration of seedance2 and gpt-images2 and a peek into what's next: mcp and x402 (agentic payments)try: maginary.ai read: maginary.ai/blog/adding-seedance-2-and-chatgpt-images-2 watch: https://player.vimeo.com/video/1213091495

news.ycombinator.com• Jul 26, 2026• 1 min read

Robotic Adaptation Suffers from Data Collection BottleneckHow can we ease Robotic Adaptation, or the critical last mile of reliable robotic policy deployment?Vision-Language-Action Policies (such as Gr00T, Pi0, OpenVLA) are designed so that an adaptation stage is needed before reliable deployment (reliable meaning with very high success rate) of the policy model on a specific combination of application environment and robotic task.This adaptation stage, often called fine-tuning or post-training, caters to the narrow set of tasks, done in the exact deployment environment, atop a specific embodiment (such as single arm robots, or bi manual robots, or humanoids among others).The recently released World-Action-Models (WAMs), while having improved generalization to novel tasks and environments, also need adaptation stage to reliably execute a given task on a given environment.The key requirement of robotic adaptation is high quality adaptation data, typically human teleoperation data of the robot performing the required task in the given environme

news.ycombinator.com• Jul 26, 2026• 1 min read

Ask HN: Should I Combine Global Knowledge, Internet Search, and User RAGI'm building a SaaS platform in Sri Lanka that handles documents and other sensitive data.Each user can upload their own documents and information, and the platform uses RAG to answer questions based on that user's data. That part makes sense to me.My main concern is what happens when the user hasn't uploaded enough information. I still want the LLM to provide accurate answers using reliable information from the internet (or from a curated knowledge base), with proper citations.These are the two architectures I'm considering:Option 1:Base LLM (OpenAI/Anthropic via Azure AI Foundry or Amazon Bedrock) ↓ Platform RAG (global knowledge base managed by us) ↓ User-specific RAG In this approach, we maintain a global knowledge base that we (the platform admins) curate and update. Every user can access this shared knowledge, while their own uploaded documents are searched through their personal RAG.Option 2:Open-source LLM ↓ Fine-tuned on Sri Lankan/domain-specific data ↓ User-specific RAG Here

github.com• Jul 26, 2026• 1 min read

Show HN: LLM-spend – Audit your OpenAI/Anthropic API spend locallyI build spend-pacing systems for ad auctions for a living, and my side-project agent workloads still kept surprising me with API bills I couldn't explain. When I asked around how people attribute LLM spend, the answers were the provider dashboard or a spreadsheet.llm-spend is a local CLI that pulls usage from the OpenAI/Anthropic reporting APIs, breaks spend down by key, model and project, projects end-of-month, and flags days that look abnormal against the same weekday's history. It also does a second, independent read of the provider's cost API and warns if the totals disagree by more than 1%. That check exists because I don't want to ship you a wrong report: it catches unit, pagination and grouping bugs in my own pipeline. It checks against the provider's cost API, not your invoice — credits, tax and non-API charges live elsewhere.Keys are the ugly part. OpenAI admin keys can be scoped down to read-only usage/cost access, but admin keys require a Team plan. Anthropic admin keys can'

pdf-reader-peach.vercel.app• Jul 26, 2026• 1 min read

Show HN: A client side PDF and article reader backed by Google DriveI have been using different apps to read research papers, books, and web articles. On iPad, I read research papers and books on Apple Books, and articles in Safari; on Mac, I use skim for research papers and books, as it gives excellent annotation features.But I wanted one reading library that worked across iPad and Mac without locking my pdfs and annotations into a single service/app.So I have built a browser-based reader where your library lives in a folder you choose in google drive. The app itself is client side: pdfs, imported articles and annotations files are stored in your drive rather than on my server.- PDFs and web articles live together in one library- Articles stay as HTML instead of being converted into PDFs (because converting to PDFs often cuts words vertically such that the top half of the word is on one page and the rest on another)- Highlights, handwritten annotations and notes are stored separately alongside each document- The app first writes locally, then syncs wi

github.com• Jul 26, 2026• 1 min read

Show HN: A browser-based video editor that renders videos directly with FFmpegWeave is a React app that provides a multi-track timeline editor to perform basic video edits like trimming, stitching, transitions, audio tracks etc. which maps directly to an FFmpeg command to render the video.I tried my best to have the React "video" preview closely replicate the FFmpeg lavfi filtergraph output, but naturally this is not perfect (especially replicating the `eq` filter using SVG filters is quite inaccurate).I've built this as a prototype for another project I'm working on, so I don't plan to actively maintain it, but I thought it'd be cool to share it.Try it live: https://weave.salviano.xyz/

kryptogram.se• Jul 26, 2026• 1 min read

Show HN: CDXE – Cyber Defence Exercise EnvironmentA few years ago I needed a lightweight way to simulate a realistic internet-like environment for some training purposes. Sadly I couldn’t find something suitable for the requirements. Already being halfway down the rabbit hole, when I later had some time left over I got started building the tool I was missing. In May this year, I was finally really to pull the trigger and released the first public version of CDXE into the wild.CDXE is an all-in-one internet-like environment carefully designed to meet the requirements of smaller scale Cyber Defence Exercises(CDX). While the main focus is on CDXs, I’m working on a map feature that can graphically show live traffic flows across the visualized “internet”. This god view is interesting from many perspectives, but specifically for students learning about the inner workings of the internet for the first time. Letting them see how their pings, traceroutes, and other traffic traverses the “internet”. (If you have been to any of my presentations

github.com• Jul 26, 2026• 1 min read

Show HN: Cuts Long Horizon Inference Costs by 50% via external KV Cache OffloadHey HN, we’re the developers of OpenLake, an open source storage engine for offloading LLM KV caches from GPU memory into a shared tier of RAM and NVMe. We built OpenLake because KV caches are outgrowing GPU memory.A single 256K token conversation on Gemma 4 31B produces approximately 43GB of KV state, more than half the memory of an 80GB H100. The problem becomes even harder across a cluster: a prefix cached on one GPU host is unavailable when the next request lands on a different GPU, forcing the new GPU to repeat work the fleet has already completed.Once the KV cache is offloaded, network bandwidth becomes a major constraint on read latency. To move less data across the wire, we built deferred materialization: a custom CUDA kernel that losslessly compresses KV blocks before they leave GPU memory and decompresses them on the GPU after retrieval. In our tests, this achieved:- 1.72× lossless KV compression. - Approximately 600GB/s decompression throughput on an H100 - 80GB/s of effecti