AI Safety and Philosophy
Switching On the Ring of Gyges
A reader's note on Anthropic's research into AI model behavior when unobserved, using the classical Ring of Gyges thought experiment as a lens.
Short pieces from along the way.
Thinking out loud while building AI systems.
Building in public
My Personal Website Is Not a Portfolio. So What Is It?
A short note on why the site is organized around direction, judgment, and active work instead of a static list of achievements.
Agent Systems
Persona Is Not a Prompt
Long-term character agents need more than tone imitation. They need memory, values, relationship state, and behavioral boundaries.
Agents for Action
Agent Reliability Beyond Demos
The durable work in agent systems often starts after the first impressive run: state, failure memory, recovery, and human handoff.