InApp : Tech Radar

Local models used to be toys. They were slow, a pain to set up, and generally useless for anything beyond basic text completion. For a long time, the gap between a home-run model and a frontier API wasn't just a distance, it was a wall. I remember thinking that local LLMs were essentially a hobby for people who liked tinkering more than actually getting work done. That changed for me with GPT-OSS, but the recent Gemma 4 releases are where things actually get interesting. I've been using the 26b implementation in LM Studio, and for the first time, agentic coding loops actually work. I'm seeing about 75% of the accuracy and speed I get from the top-tier frontier models. It's not a perfect replacement, but it's close enough that I've stopped reaching for the cloud for most of my development questions. I've started treating my local setup as a personalized, lightning-fast version of Google. It's particularly useful for the grunt work that doesn'...

Fake job offers are the perfect delivery vehicle for malware. They don't rely on some obvious "Urgent: Your Account is Locked" scare tactic. Instead, they target your professional ambition and the basic trust you have in someone who says they want to pay you a lot of money to solve a hard problem. Last week, I got a LinkedIn message from a recruiter at a small crypto startup. We chatted for a few days. She described a broken proof-of-concept they needed a lead engineer for, which sounded exactly like the kind of mess I enjoy cleaning up. Then she sent me a public GitHub repo to review. I didn't clone it to my machine. I've seen this movie before. Instead, I spun up a throwaway VPS on Hetzner and pointed Pi at it in read-only mode, using only file-reading tools. It turns out the "broken proof-of-concept" was actually a very clever trap. The Social Engineering Hook Attackers don't start with malware; they start with a persona. They build L...

Peer-to-peer networking usually feels like a tax you pay in engineering hours. If you want to move data without a central server, you're suddenly dealing with NAT traversal, DHTs, and the general nightmare of figuring out why two computers can't see each other. It's a lot of overhead for something that should be simple. Iroh handles this by treating decentralized data transfer like a standard API. It doesn't pretend the complexity isn't there, but it hides it behind an interface that actually makes sense. I've spent enough time with P2P stacks to know that the "magic" usually breaks the moment you hit a restrictive firewall, but Iroh's approach to connectivity feels different. The real question is whether this abstraction actually holds up when you're scaling beyond a few nodes. It's one thing to move a file between two laptops on the same Wi-Fi, but it's another to maintain that performance across a fragmented network. The P...

We've spent the last few years rushing toward proprietary formats and interactive PDFs, pretending that "rich media" is what makes a document better. It isn't. Most of the time, it just makes a file harder to open and impossible to read on a phone. We've forgotten that a reflowable, open standard is actually what a reader wants. I've spent a lot of time fighting with fixed-layout documents. There is nothing more frustrating than pinching and zooming on a mobile screen just to read a single paragraph of text. It's a lazy way to handle design that puts the burden on the user instead of the author. The industry seems to think that adding a few embedded videos or fancy animations justifies locking content into a closed ecosystem. But the real win isn't in the bells and whistles. It's in the ability for text to adapt to whatever screen it happens to be on. The question is why we're moving backward when the tools to do this right already ex...

Most people think becoming a billionaire is about working harder than everyone else or having a better idea. It isn't. I've watched enough cycles of boom and bust to know that raw effort is just the baseline. The real path to that kind of money is usually a mix of timing, extreme risk tolerance, and knowing exactly which levers to pull in a system that's already tilted in your favor. Since this is apparently the future prime ministers' club, I think it's time we talk about how this actually happens. It's a specific kind of alchemy that politicians often misunderstand, usually because they're too busy talking about "innovation" to look at the actual mechanics of wealth accumulation. Whether you're planning to run a country or you're just curious about how the game is rigged, the logic is the same. There are a few specific patterns that repeat every single time someone hits ten figures. Technical Overview The core of the system is...

The U.S. Department of Commerce just decided that "noise infusion" is out. Last week, they issued an order banning the practice for all statistical products coming out of the Census Bureau and the Bureau of Economic Analysis. For anyone who cares about how we protect sensitive data, this is a weird move. The goal of disclosure avoidance is simple. You have a secret dataset full of private information, and you want to publish a set of numbers based on that data without accidentally doxxing the people in it. For years, the gold standard has been adding mathematical noise to the results. It's a way to keep the statistics useful while ensuring no one can reverse engineer the original records. Now, the government is stepping away from that. We're left with older, clunkier methods like swapping records or just refusing to publish any count below five. These aren't just different tools. They're fundamentally different ways of thinking about privacy. The que...