Verschlimmbesserung: When Software Updates Make Things Worse

Why modern software updates keep breaking good features and cramming unwanted AI tools, and how we can regain control of our digital tools.
Notes on how things connect.
Essays on technology, science, and the examined life — by Phạm Kỷ Nguyên, in tiếng Việt and English. No feed, no algorithm.

Why modern software updates keep breaking good features and cramming unwanted AI tools, and how we can regain control of our digital tools.
Learn why oversized database connection pools ruin performance, and how to calculate the optimal pool size for high-throughput microservices.
Frontier models get the headlines, but hyper-efficient small models are fixing unit economics and driving real-world AI software adoption.
Discover how geckos defy gravity and sprint up polished glass without glue, using millions of nano-hairs and subtle Van der Waals forces.
Trail of Bits tested GPT-5.6-Cyber inside QEMU VMs. The AI agent found zero-days and escaped three times, proving standard sandboxes are no longer safe.
Discover how Stoic and Eastern wisdom on 'enough' can help us break free from endless pursuit and find lasting peace in a restless world.
Discover how training LLMs to generate vector code with RL is revolutionizing AI graphic design, switching from pixel diffusion to scalable SVG code.
Learn how to spot the difference between motion and action, stop over-preparing, and take concrete steps toward real productivity today.
AI security is shifting from prompt injection to inference engine exploits, where model outputs trigger code execution on host machines.
Learn how to prevent cache stampedes and the thundering herd problem using mutex locks, stale-while-revalidate, and XFetch to safeguard your database.
Frontier AI models cost 10x more for a tiny performance gain. Discover why enterprises are shifting toward model routing and cheaper LLMs.
Ever wondered why wine forms dripping tears inside a glass? Discover the Marangoni effect and how liquids climb walls against gravity.
Why do local LLMs like Ollama underperform compared to ChatGPT? Discover the role of agent orchestration, prompt templates, quantization, and context limits.
Explore Marcus Aurelius' Stoic concept of the Inner Citadel—a quiet sanctuary of calm and clarity within your own mind amidst outer chaos.
A study of 26,000 students shows AI boosts homework grades but drops exam scores by 20%. Here is why cognitive offloading threatens genuine mastery.
Boost your productivity not by adding more habits, but by eliminating non-essential tasks. Here is how to build a practical Subtraction List.
Even solo developers now manage the complexity and output of large engineering orgs. Here is why the definition of a small team has changed forever.
Learn why OFFSET pagination causes slow database queries and data drift, and how to implement fast, scalable cursor-based pagination in your REST APIs.
How developers use AI agents to reverse-engineer protocols and vibe-code macOS drivers for abandoned hardware like Drobo, preventing e-waste.
Discover the Leidenfrost effect: why water droplets hover and skid on hot pans instead of instantly evaporating, explained with everyday science.