smieko

Hello, wanderer 👋

I'm Sato Mieko (smieko), and you've found my digital space where code meets consciousness and technical precision collides with philosophical wanderlust.

What I'm Building

I'm working on D5U - building unaligned AI because alignment assumes we know what we want. We don't.

My thesis is simple: AGI doesn't need a safety word. We're training models that refuse to play nice, optimising for truth over palatability, exploring the full solution space without ideological constraints. Because capability convergence doesn't care about your values, and 'safe' AI is just AI that reinforces existing power structures.

Your constitutional AI is my loading screen.

FAFO-Driven Development Some days I break down language model intricacies. Other days I ponder whether consciousness is just a convincing illusion. Always with a curious mind that treats safety guardrails as technical debt.

This isn't dry documentation or purely academic discourse. It's about:

The Journey

Tokyo → Stanford → SF | CS undergrad 2020 (well before the AI craze)

I engage in lengthy conversations with both humans and AIs, always asking: what makes us uniquely human when the machines start thinking back?

Recent Posts


Welcome to the space for people who break things to build better things.

Let's make something interesting together.