A point cloud, shaped like a brain, generated in code rather than modeled or rendered from a stock file. It turns slowly. The bright nodes are not decoration: each one is a real area of focus, drawn from what I actually work on and read right now. Drag to look from another angle. Hover or tap a node for what it holds.
ドラッグで回転、ノードにホバーまたはタップ。
上のノードにカーソルを合わせるか、リストから選んでください。
専門分野
考えていること
読んでいる本
すべてを貫く一本の糸
この部屋について
The shape is procedural, not a medical scan: two overlapping point-clouds stand in for the hemispheres, a folded surface for gyri, generated live in your browser. The content is real, the same public "Digital Brain" material from the current site: areas of expertise, current reading, current thinking, the pull quote, just given a better home.
マウスを使わない方のための全ノード一覧
AIガバナンス: How a system earns the right to be trusted, not just the right to be used.
機械学習: The craft underneath the claims. I want to know what a model actually did, not just what it says it did.
規制: Rules are only as good as the evidence a regulator can independently check.
リーダーシップ: Building a company around a hard technical bet, and staying honest about what is proven versus not yet.
検証可能なAI出力: What it would take for any output to carry its own proof of how it was produced.
LLMのハルシネーション低減: Less a bug to patch than a property to measure, honestly, every time.
EU AI法対応ツール: Turning a legal deadline into something a small team can actually build against.
エージェント型AIの未来: Systems that act on their own need an audit trail even more than systems that only answer.
The Alignment Problem: Brian Christian. Where the machine-learning story and the ethics story stop being two separate books.
Human Compatible: Stuart Russell. The argument for building systems that stay uncertain about what we want, on purpose.
AI 2041: Kai-Fu Lee. Ten near-future scenarios, useful precisely because most of them are already starting to happen.
核心となる引用: The future belongs to those who understand that trust in AI must be built, not assumed.