The latest blog posts to take off on Hacker News. 1,423 posts.
Walk through Karpathy's 200-line GPT from scratch. Tokenize names into integers, watch softmax convert scores to probabilities, step through backpropagation on a computation graph, explore attention heatmaps, and see a tiny model learn to generate plausible names.
Why is my first C++ (m)allocation always 72 KB?? It’s the ’emergency pool’ for exceptions, here’s how I uncovered it using GDB and digging through the source code of libstdc++, glibc and Valgrind!
How Replit turns any web page into a deterministic video by virtualizing time itself, patching key browser audio APIs, and waging war against headless Chrome's quirks.
Neural networks are increasingly deployed through software stacks whose behavior is defined by tensor layouts, preprocessing choices, compiler boundaries, floating-point arithmetic, GPU kernels, and external model exporters. TorchLean is a Lean 4 codebase for neural network specification, execution, and verification…
A systems analysis of how AI-assisted development creates a gap between output speed and understanding, and why organizations cannot see it happening.
In my previous post I talk about how I got rid of hundreds of thousands of lines of Objective-C code while at Audible and I explain why keeping Objective-C code around is a terrible idea. And I explain that…
A programming language should not be your identity.
Here's an interesting bug I ran into recently: After transferring some files to my server via scp, I couldn't log in via SSH anymore - my key was rejected by sshd.
Comparing a career flying the 747 and a career coding with AI assistance.
The Web streams API has become ubiquitous in JavaScript runtimes but was designed for a different era. Here's what a modern streaming API could (should?) look like.
Benchmarking and Ground Control for developer tool providers. Understand agent choices and test what helps them succeed.
Is your vibe-coded app a Crapject or a Gift?
Today, it’s my great pleasure to announce the first public release of Hardwood, a new parser for the Apache Parquet file format, optimized for minimal dependencies and great performance. Hardwood is …
When a codebase is large and complex, it's risky to big changes. Here's a structured way to approach
Google spent over a decade telling developers that Google API keys (like those used in Maps, Firebase, etc.) are not secrets. But that's no longer true.
AI agent misbehavior isn't a sandbox problem—it's a permissions problem.
Hello Windows Insiders, today we are beginning to roll out updates for Notepad and Paint apps to Windows Insiders in the Canary and Dev Channels on Windows 11. Notepad (version 11.2512.10.0) In this update, we’re introduci
_posted by Stephen De Gabrielle and John Clements_ We are pleased to announce Racket v9.1 is now available from https://download.racket-lang.org/. As of this release:: Documentation organization and navigation can be specialized by language family, to...
confusable-vision renders every TR39 confusable pair across 230 macOS system fonts and measures visual similarity with SSIM. 96.5% of confusables.txt is not high-risk, but 82 pairs are pixel-identical in at least one font.
From COBOL in the 1960s to AI in the 2020s, every generation promises to eliminate programmers. Explore the recurring cycles of software simplification hype.
An interactive exploration of quadtrees. Start with brute-force spatial search, build recursive space partitioning step by step, then use it for point lookups, range queries, nearest-neighbor search, collision detection, and image compression.
I'm building an 80386-compatible core in SystemVerilog and blogging the process. In the previous post , we looked at how the 386 reuses one barrel shifter for all shift and rotate instructions. This time we move from real mode to protected and talk about protection .
The author of Java 26's MemoryMXBean.getTotalGcCpuTime() analyzes the CPU-memory trade-off and why pause times fail to capture explicit GC cost.
Wolfram's Foundation Tool supplements LLM foundation models with deep computation and precise knowledge. CAG (computation-augmented generation) underlies multiple access methods.
Writing code is cheap now - Agentic Engineering Patterns
Righty-ho, I’m back from Rust Nation, and busily horrifying my teenage daughter with my (admittedly atrocious) attempts at doing an English accent 1 . It was a great trip with a lot of good conversations and some interesting observations. I am going to try to blog about some of them, starting with some thoughts…
We trained a model on our 11-million-hour video dataset. Our model can explore complex websites, complete multi-action CAD modeling sequences, and drive a car in the real world, all at 30 FPS.
We're adopting Rust as our C++ successor language, and using AI agents to accelerate the transition.
Most linux software just work on FreeBSD
