Tech Radar

Moonshot AI just dropped Kimi-K3 on HuggingFace with a 3 trillion parameter count. On paper, that's a massive number. In reality, it's a nightmare for anyone without a literal warehouse of H100s. I'm not sure why we're still treating parameter counts like a high score in a video game. A bigger model doesn't automatically mean a smarter one, but it always means more VRAM overhead. If you're trying to run this on anything less than a high-end cluster, you're going to hit a wall pretty quickly. The real question is whether the marginal gains in reasoning actually justify the hardware cost. I spent the morning digging into the weights to see if the performance jump is real or if we're just seeing the limits of brute force. The Scale of Kimi-K3 Kimi-K3 is a 3 trillion parameter model. Seeing a model of this size on HuggingFace is rare, mostly because the hardware requirements are brutal. It uses native mxfp4 quantization to keep the memory footpri...

Most people treat their database like a black box. You send a query into the void, and eventually, some rows come back. If it's slow, you add an index or throw more RAM at the instance and hope for the best. But that's a lazy way to manage infrastructure. I think it's more helpful to view the PostgreSQL engine as a coordinated city. You've got processes acting like traffic, memory buffers behaving like warehouses, and the write-ahead log acting as the city's permanent record. When you stop guessing and start visualizing how these pieces actually move, the "magic" of query optimization disappears and is replaced by actual logic. There's a project called PGSimCity that turns this mental model into a literal simulation. It's a weird, visual way to see what's actually happening under the hood of your cluster. It makes me wonder why we've spent decades staring at text-based logs when the engine is doing something this rhythmic. The Archite...

A US citizen is facing charges because their phone wiped itself during an airport security check. The device was running GrapheneOS, an open-source project that lets you trigger a full data wipe if you enter a specific passcode. It's a feature designed for people who care about privacy, but the legal system is treating it like an act of obstruction. I've followed a lot of privacy-focused OS projects over the years, and this feels different. Usually, the tension is between the user and the state. Here, the law is effectively targeting the operating system itself. It's a weird, uncomfortable precedent. We're seeing a genuine gap between what's technically possible for privacy and what's legally permissible. If using a specific piece of software to protect your data is now a liability, where does that leave the rest of us? The Incident The device wipe happened because the user triggered a remote erase command while the phone was in custody at the airpor...

HyperCard was a weird, beautiful anomaly. It treated everything as an object, blurring the line between the tool you used to build a project and the project itself. For a long time, that philosophy felt like a relic of the 90s, something we traded for the rigid separation of "development" and "production." Decker brings that mindset back to the modern web. It's a multimedia platform for interactive documents. You get sound, images, and scripted behavior, all running in your browser. I've spent a few hours with it, and the most interesting part isn't the media support, but how it handles widgets. You can define a custom interface element and then just copy and paste the definition using your system clipboard. That's a strange, tactile way to share code. Instead of pushing a package to a registry or linking a GitHub gist, you're just moving a piece of a living document from one place to another. It makes the act of building feel less like ...

Google just let slip in a routine financial disclosure that they've been quietly hoarding a massive stake in SpaceX. It's the kind of detail that usually gets buried in the footnotes, but the scale of it is jarring. I can only hope they locked in this investment years ago at a fraction of the current valuation. If they bought in recently, this is a disaster waiting to happen. I'm short on both of them right now. SpaceX's eventual IPO smells foul to me, and Google is essentially a giant AI-bubble short while Anthropic and OpenAI continue to sit on the sidelines. The real question is whether Google is actually betting on the future of aerospace, or if they're just doubling down on a valuation that's drifted too far from reality. The $94.1 Billion Disclosure Google's filing shows a $94.1 billion loss, which is an absurd number that's easy to misread. This isn't a typical operational loss where they spent more on servers than they made in ad...

Cookie banners are a complete failure. They've become this universal tax on the internet that manages to annoy every single user while providing a false sense of legal security for the companies that implement them. We've all just conditioned ourselves to click "Accept All" without looking, which means the banners aren't actually protecting anyone's privacy. They're just digital noise. Most people assume these pop-ups are a strict requirement of EU privacy law. They aren't. The law is actually pretty clear: online tracking is prohibited by default. The banners are just a clumsy workaround that companies use to try and trick us into opting in. The EU Commission finally proposed a way out of this mess. Instead of playing whack-a-mole with every single website you visit, you'd set your privacy preferences in your browser once. That's it. No more banners, no more fake choices. Whether this actually kills the cookie pop-up for good depends ...