CAPITAL IDEAS: Can small-cap stocks continue to outperform? - The Berkshire EdgeCAPITAL IDEAS: Can small-cap stocks continue to outperform? The Berkshire Edge
Show HN: Sprocket – The Best AI Agent for Hardware and Software DevelopmentHey HN, I am 16y/o and have been working on Sprocket for a while. It's an open-source AI agent that beats every other agent out there at both hardware and software.And here's the best part: Sprocket can (on its own) buy anything from any website when you tell it to do so. From hardware parts to SaaS subscriptions.Sprocket retrieves best-in-class context from the web for everything it does. It is therefore incredibly reliable. The agent harness's quality, performance, and UI rival that of Codex/Cursor/T3Code, and we are rapidly improving. We will be releasing benchmarks on how Sprocket beats every other agent at both software and hardware soon.In terms of hardware design, Sprocket can make beautiful schematics in react, create your BOM, and create detailed assembly instructions.Hope you try it out!Thanks, AaravP.S. if the demo video doesn't load for you, come back after ~1.5hr and it should be there, I kinda had to submit this early. Sorry!
Show HN: Rextio – Compile Python code natively using Rust with CPython fallbacksHi, HN! Let me introduce my experimental project Rextio.While Python is widely used in fields such as AI and data analysis and offers the advantage of fast coding, it often causes some performance issues because Python is an interpreter language.Of course, there are various alternatives to address Python's speed issues. In some cases, Python code is used only for initial prototypes, and the development is re-evaluated in Rust or C/C++ upon transitioning to production.However, I'm experimentally developing this tool called Rextio as a new alternative to solve speed problems without abandoning Python.Rextio improves performance by automatically converting parts of Python hot paths that can be compiled natively (such as type-hinted static code) into Rust code and compiling them into binaries. It then falls back to the rest of the code as is and executes it using CPython. This allows for the complete preservation of the original Python logic while aiming to improve execution speed for spec
Logitech vs. Razer: Which brand makes better gaming gear?We may receive a commission on purchases made from links. Gamers are passionate about their play. They're constantly seeking ways to improve the gaming experience or optimize their performance. And ...
Show HN: Open-Cowork – an open-source, model-agnostic computer-use agentHi HN,We built Open Cowork, an MIT-licensed desktop agent that can operate a computer through screenshots, mouse input, and keyboard input.We started working on this after using cowork-style computer agents for longer tasks. The interaction model was useful, but computer use burns through model calls much faster than chat: take a screenshot, decide what to do, perform an action, inspect the result, and repeat.We wanted to separate that agent loop from any one model or subscription.Open Cowork currently works with local models through Ollama and hosted models through providers including OpenAI, Anthropic, Google, xAI, Mistral, and OpenRouter. You can watch the agent’s actions as they happen and interrupt it during a run.The basic loop is:Capture the current screen Send the screenshot and task state to the model Parse the proposed action Execute the mouse or keyboard action Capture the resulting state Continue until completion, interruption, or a safety conditionTwo annoyingly common fai
Kimi-code not performing wellI managed to get a subscription to kimi and have been running kimi-code for the last few hours, having imported my claude-code skills, settings etc into it.I pointed it at a JS and python codebase I have been working on and asked it to review and refactor, then I got it to add some features - and boy, I am not impressed.TBH it looks the part with the running commentary it makes, but it seems to make some dumb mistakes.Compared to claude-code it is slow, time consuming, hungry on the tokens and multiple times it told me it was done and the app could not even start.I explicitly, and without any ambiguity, have standing orders that there is no type of silent failure that is acceptable in the current foreseen lifetime of our universe, and it still embedded a bunch of them in what was a relatively small amount of code I got it to generate.I am just not seeing that it comes close to claude-code running Anthropic models, based on this one particular codebase I have tried it on so far.
Show HN: DispatchSEO – open-source SEO manager tool for Claude CodeThis SEO tool is similar to traditional ones: Semrush, SEObot and OutrankKey features:- research keywords- analyze rankings- create content workflows- monitor search performanceTech stack:- NextJS 16 (React 19, App Router)- TypeScript- Tailwind v4- PostgreSQL (bundled container self-hosted, Supabase in cloud)- MCP server (mcp-handler + u/modelcontextprotocol/sdk)- Claude Code, headless, in a container (the builder)- Docker Compose (app + db + cron + builder)- GitHub Actions (schedules and PRs on your repo)- Google Search Console API + DataForSEO + SerpApi- Resend (email alerts)Fully open-source (AGPL-3.0)
9 reloading presses that serious handloaders say outperform everything costing twice as muchSome presses look impressive in a catalog. These are the ones reloaders actually keep bolted to the bench for years. Forster Co-Ax Is Still the Precision Benchmark If you spend any time around ...
New Hampshire outperformed the rest of New England in U.S. News’ ‘Best States’ ranking. Here’s how. - The Boston GlobeNew Hampshire outperformed the rest of New England in U.S. News’ ‘Best States’ ranking. Here’s how. The Boston Globe
Quantum computers outperform classical ones, with results you can trustThere are many algorithms for which it has been mathematically proven that a quantum computer can generate results that would take a classical computer an unreasonable amount of time to generate.
The World Is to Focused on Quarterly EarningsEverybody seems to care way too much about what happens in a quarter- nothing material usually, but people make a huge deal about it.There are two camps in investing: those that think there is a bubble and those that think everything always goes up. There both wrong. Bubbles are inherent with a new wave of technology; each one is different and has its own circumstances. You should not wait for anything because you anticipate one. You cannot time the market well; you should focus on finding companies that either are in the process of developing a moat or adding or transitioning business models, but have not done any of these things yet. You have to analytically figure this out. There is no way to outperform the index by investing in AAPL because you think they have a great brand; so does everyone else.The camp that thinks everything goes up is right to criticize the people that think there is a bubble because memory sold off, or a company had bad earnings. Technically, no fundamentals h
University of San FranciscoUniversity of San Francisco 2026 Law & Specialties Rankings University of San Francisco is ranked No. 156 (tie) out of 194 in Best Law Schools. Schools were assessed on their performance across a set ...
Show HN: A new engine to run Kimi K3 on a laptopKimi K3 has 2.78 trillion parameters and ships as 1.42 TB of weights. It clearly does not fit in the memory of a laptop. But K3 is a Mixture-of-Experts model. For each token, only a small fraction of its 896 experts per layer is activated. That changes the problem: the entire model does not need to be resident in RAM, as long as the weights required by each token can be reached quickly enough.We built WASTE — the Weight-Aware Streaming Tensor Engine — to explore that idea.WASTE keeps the dense, repeatedly used part of the model resident in memory, stores the routed experts in an NVMe-optimized container, and streams only the experts selected during inference. The remaining RAM is used as a bounded expert cache.The current Kimi K3 container is 982 GiB. On a 64 GB MacBook Pro, WASTE runs the complete model at around 0.32–0.34 tokens per second, with a measured minimum memory requirement of approximately 29 GB at a 4K context.That is obviously not interactive performance yet. But the resu
Raiders new head coach Klint Kubiak names Kirk Cousins starting QB over No. 1 pick Fernando MendozaKirk Cousins of the Las Vegas Raiders practices during an OTA offseason workout at the Las Vegas Raiders Headquarters/Intermountain Health Performance Center on June ...
Show HN: Purple Hammer – construction takeoff and estimating in Tauri/Rust/WASMPurple Hammer does construction takeoff. You measure quantities off PDF drawings and a built-in spreadsheet lets you read from databases, formulas, assemblies etc to price the job. This matters because most takeoff software (Bluebeam, Planswift) do not have the spreadsheet side of things built in.Stack is Tauri 2, Rust, SolidJS. The render and spreadsheet engines are both custom, compiled to WASM with modified PDFium for parsing. One codebase builds native Mac and Windows.Performance is most of the work. A 500-sheet set opens in about a second. Million+ vector drawings hold 120 FPS on mid hardware. It feels somewhat faster that any other construction viewer I’ve tested. Compare the live demo (on the homepage, or download free trial) vs an industry standard pdf tool at showcase.apryse.com/cad-viewer. The solver runs 11-15M dependency steps/sec on an adversarial test.Perhaps most important engineering achievement is that RAM & Performance scale strictly with active attention rather than
Show HN: Cosmos47, a public chronological feed without algorithmsI’ve worked as a B2C product manager and co-founded two social-media-related startups (one acquired and one closed down), but I was never particularly good at conventional social networking. I don’t enjoy cultivating followers, optimizing for engagement, or wondering how a post will perform.Yet sometimes I want to share something and I want people to get a reasonable chance of finding it. Sure, I can start a blog or an X account (which obviously I have) but when I post something, usually nobody sees it or interacts with it.So... I've built Cosmos47 - a public social feed that shows all posts chronologically. One feed that appears the same to all users, from all over the world. No algorithm or ranking people by popularity.All posts are analyzed with AI and each post gets tagged with its language, topics, tone, format, length, location, and other characteristics. Readers can then filter the feed themselves (if they wish). The idea is distribution through relevance rather than popularity.
Microsoft touts cost-saving AI model for cybersecurityMicrosoft talked up its first artificial intelligence model for finding risky parts of source code. When connected with OpenAI's general-purpose GPT 5.4, Microsoft says the new model can outperform ...