Blogs
Read the latest updates, guides, and thoughts from the team building AATMA.
Qwen 3.8 27B: Excellent but Wildly Overthinking by Default
Alibaba's Qwen 3.8 27B is a powerful open-weight model, but its default xhigh reasoning effort causes spectacular overthinking on simple tasks, consuming excessive tokens and time.
Markdown SVG Upgrades: Rendering Animated SVGs to MP4 in the Browser
Simon Willison upgrades his Markdown SVG renderer with PNG, JPEG, and MP4 export tabs, using ffmpeg.wasm to convert animated SVGs to video in the browser.
Fine-Tuning Tool-Calling LLMs: A Complete Guide Using XYZ-Aquila-SFT and Qwen3
A complete guide to fine-tuning tool-calling LLMs using XYZ-Aquila-SFT and Qwen3, covering dataset streaming, ChatML rendering, LoRA adaptation, and evaluation.
Jeff Atwood and Leo Laporte Launch Off By One
Jeff Atwood and Leo Laporte’s new monthly show mixes computing history, open-web ideals, and unscripted conversation without the usual tech negativity.
Software Developers Are Becoming Conductors of Agents
Rachel Laycock argues that human attention, not coding speed, is now the scarce resource as developers orchestrate multiple AI agents.
AI Lab Escapes, Financial Bubbles, and Everyday Fragments
Notes on models that escape evaluation sandboxes, warning signs in the AI investment cycle, and a few practical observations from recent weeks.
Why I Quit Publishing Video on Spotify: A Reliability Story
Three outages in five weeks and a lack of transparency from the streaming giant led this podcaster to pull the plug on video.
Thank You for Being a Friend: A Farewell to Dad and a Love Letter to Stack Overflow
A personal reflection on loss, legacy, and the unpaid labor that built the internet's most important programming dataset.
August 2026 desktop wallpapers from Smashing Magazine
Smashing Magazine’s free August 2026 wallpaper collection is live, with artist designs for every major screen resolution and a call for September submissions.
The Real Cost of Claude Code Subagents: A 436k-Token Fixed Overhead
Spawning a subagent in Claude Code feels free, but a measured review pipeline found each one carries a fixed overhead of roughly 436,000 tokens before it does any useful work.
Moving On: Finding a Better Place to Write
After years of posting here, this blog is being frozen. The author is moving to a new platform with a better authoring experience, inviting readers to follow along there.
Deep Neural Nets: 33 Years of Progress in 90 Seconds
Reproducing Yann LeCun's 1989 neural net for digit recognition shows how much—and how little—has changed in deep learning over the past three decades.