Tech Stack Blog
Notes on tech stacks and architecture decisions
All About RAM (8) — Micron Says Memory Supply Gets Tighter Through 2028
Micron's supply outlook from its earnings call, the 2027 volume already under contract, and what customer prepayments say about a multi-year rise in memory prices
How Much Space Apple Intelligence Takes on macOS 27 — 19.33GB on an M3 Pro
The real size of Apple Intelligence model files that don't show up as their own line in Storage, how to check it, and why they come back after you delete them
Seven Korean Financial Firms Breached in a Row — The Entry Point Was Staff-Facing Systems
How the breaches that started at Shinhan Bank spread across the financial sector in five days, the common way in, and what customers can do
Claude Sonnet 5.5 Released — Scores Close to Opus 5.5 at Half the Price
Benchmarks within 1–2 points of Opus 5.5, unchanged pricing, and how to choose between the two models by task
Aleph Alpha Kolibri — German Sovereign AI Released Under Apache 2.0
The makeup and performance of Kolibri, a 78B MoE model built and released in Germany, and how its direction differs from Korea's sovereign AI foundation model program
Galaxy Buds On — Samsung's First Clip-On Earbuds at ₩299,000
Specs and launch schedule of the open-ear, clip-style Galaxy Buds On, the sound-leakage problem, and how competing products are priced
Two Citrix NetScaler Zero-Days — Updating Doesn't Remove the Web Shells
How two NetScaler remote code execution flaws exploited before disclosure unfolded, and why you should check for signs of compromise before you update
Dead-Simple RAG (2) — LoRA Fine-Tuning on Court Rulings Hits Half as Many Statutes as RAG
A model LoRA-fine-tuned on the same court-ruling corpus learned only the format, and fell short of RAG at citing the right statutes
OpenAI dots — An Agent That Works 24/7 on Its Own Cloud Computer
How dots, the always-on agent that keeps working after the conversation ends, is put together, where it's available in Korea, and what it can and can't be trusted to do
Gemini 4 Argon Unveiled — 1M-Token Output Limit, Cyber Defenders First
The output limit, benchmarks, and introductory price of Argon, the first Gemini 4 model, and its staged release that opens to cyber defense organizations first