Introducing SWE-2: Pushing the Pareto Frontier Today we’re introducing SWE-2, our most advanced coding model yet. SWE-2 delivers highly competitive agentic coding performance across multiple effort…
Software Drives People Insane I have a little pet theory that software drives people insane. Not in the "wash your hands every thirty minutes like Howard Hughes" kind of way, but more ...
Guest Post: Rust Is Tier-1 Language at Microsoft By now it’s no surprise that Rust is of strategic importance to Microsoft. From bold mission statements when Azure CTO Mark Russinovich outlined our future strategy for native code, to millions of dollars invested by Microsoft into supporting the Rust Pro...
mr. TIM (@timkellogg.me) another researcher accuses OpenAI of training on conversations and then claiming a breakthrough
Native is now the future of mobile at Shopify (2026) - Shopify Coding agents changed what it costs to build mobile apps twice. Here’s why Shopify is moving from React Native back to Swift and Kotlin.
Streaming costs $702 a year more than in 2021 — HonestlyRanked 9 major streaming services cost $95.91/mo in 2021-03 and $154.41/mo today — up 61%, or $702 a year. Every change sourced.
WebMCP Changes How We Interact with the Web! (beginner intro with React) Use React and WebMCP to expose tools that AI agents discover and call right inside the browser. In this beginner tutorial, you'll add WebMCP to a Vite-powered React greeting app, register a typed tool, and let ChatGPT Desktop call it. You'll also see how...
They really do think AI might kill everyone A recent resignation tweet from an Anthropic researcher has everyone talking about the AI apocalypse again. Among other things, he said: The people building AI…
I think I used the Hermes Agent wrong Around three weeks ago, I made the switch from OpenClaw. The idea of a self-learning AI agent (as Nous Research advertises it) hooked me. It also matches how I think about AI agents: I want them to grow with me. I installed the Hermes Agent and gave it a ...
Getting 50 GB/s Back Out of the ANE An RTL performance erratum in the Apple M3 Neural Engine throttles DRAM weight streaming throughput down to 17–19 GB/s from the nominal 45–60 GB/s, whenever the total weight size is an integer multiple of 1 MiB, which currently affects 7 of ANEMLL’s 15 mo...
iPhone Duo: the high price is a feature Open Hacker News on any iPhone keynote day and, without fail, you'll see the same comments. Today with the launch of iPhone Duo, things were no different. "It's too big" "When are they bringing back the iPhone mini?" "It's too expensive" I can't reall...
Fileless ELF Execution via Kernel Keyring Using the Linux kernel keyring to stage an ELF in slab memory and execute it via userland exec, skipping execve and the filesystem entirely.
Apple debuts iPhone 18 Pro and iPhone 18 Pro Max Apple announced iPhone 18 Pro and iPhone 18 Pro Max, delivering an advanced pro camera system and improved battery life and performance.
Don't stick your SidePulse Pro in a Mac Studio The SidePulse Pro is a fun project that puts some LEDs in an "SD card format" to go into the side of a MacBook Pro. My MacBook Pro is very ...
What LLM coding agents have taken from me A blog containing essays, of varying cogency, about various things that my mind have wandered to.
Who Handles Your Security Reviews? LLMs are very good at finding exploits, but they're also really good at creating them — just like humans. So who handles your security reviews?
ID design and primary keys, pt. 1 Author: Alexey Makhotkin squadette@gmail.com, (~2300 words) This is the first part of a systematic discussion of primary keys in database design. As usual, we present the material in a way that deviates from the traditional approach. This is basically a b...
Building Jarvis: A Better Voice Starts Before Training Our Jarvis TTS adapter is now public on Hugging Face. Inside the work: a dataset studio for field recordings, selective cleanup, reproducible LoRA training, and the experiments that changed how we evaluate a voice.
So you want to use OpenRouter? — Mo Moustafa Might seem simple on the face of it, but unfortunately it's pain all the way down.
What makes $BigTechCompany's products bad, anyway? Companies suck for different reasons. Don't conflate all of big tech into one blob, reason over companies according to their income streams.
You Could Have Come Up with Speculative Decoding Learn how speculative decoding accelerates LLM inference through predictable tokens, rejection sampling, parallel verification, and GPU roofline analysis.
The Shift from ML Engineering to AI Engineering Exploring the key differences between ML and AI engineering, foundation models, and the challenges of building AI applications.
I asked it to stop talking down to me I think many people still do not fully understand what modern large language models are. A normal LLM communication is optimized so that you can understand it.
Measuring the noise floor of an LLM-judged eval — digline A case went from 5/5 to 2/5 with nothing changed. How I measured the noise floor of an LLM-judged eval, what it caught the week after, and where it still can't see.
LiteLLM proxy - Wirt OÜ LiteLLM proxy in front of the vLLM backends: virtual keys, health-check routing, a hardened systemd unit, and the nginx config that makes SSE streaming survive.
The new addiction to working all the time We’re all so excited about the newest capabilities of each released AI model that we’re forgetting the consequences of having access to such an amazing tool, and the possibility of being able to do so much so fast. I know people who go to the gym and stil...
(KV) Cache Rules Everything Around Me TL;DR: The biggest cost to devs of an agent is re-reading its own context, which costs the provider close to nothing to serve.
I Let Claude Merge 38 PRs. Here’s What Broke. - Craig Phares A field report on misleading green checks, durable agent state, and keeping production behind a human gate. Claude merged the sixth and final phase PR, then went idle. The plan still had one job left: open an integration PR with…
