A digital version of my messy desk. Half-finished projects, random thoughts, and life logs.
I counted which benchmarks appeared in two months of major model announcements, then sorted out what each one measures and how it scores. It also covers Artificial Analysis's combined index, the limits of benchmarks, and what carries over to evaluating your own AI systems.
A look at recent releases from Google, OpenAI, and Kimi, the model updates that led here, and how my coding setup changed along the way. I am still figuring out how to use convenient harness features without making it too expensive to switch.
I read three articles, Bun's Zig-to-Rust migration, the GitHub Copilot runtime migration to Rust, and how the Claude Code team does code review, to think about task units, the human role, and the skill of fixing loops when running agents at large scale and in parallel.
I looked into GraphRAG and LLM Wiki and found that both "distill" raw data with an LLM. Here is my understanding of how each developed, the problems they share, and which one to pick.
I wanted to make an unfamiliar codebase visible, so I looked into static analysis tools. How I use ripgrep, ast-grep, Serena and CodeQL, and what I learned from a small trial.
REA, Adobe clones, GTA and Halo in the browser. A week of "Software is over / done / dead" takes, sorted out by a software engineer: what actually ends, and what stays.
Organizing everything I learned about AI/ML during my holiday crash course. Part 1 covers general principles and an overview of classic, non-NN machine learning models.
House hunting saga in Japan. From spec-driven logic to realizing the importance of intuition.
The reality of AI development is a bit different from the magic shown on TikTok. A story about the struggle and fun of managing a digital Sheldon Cooper.
Rakuten, Moving to the US, Meta, Google, Returning to Japan, Kyndryl. Reflecting on my engineering journey so far.
From Yamaguchi, Texas-influenced, living in Tokyo.
