multi-agent 的設計判準
When should a system actually be split into multi-agent? Look first for differences that cannot be compressed back into one agent: cognition, adversarial check, attention, and work topology.
Writing, notes, and text fragments that sit beside the work archive.
When should a system actually be split into multi-agent? Look first for differences that cannot be compressed back into one agent: cognition, adversarial check, attention, and work topology.
Control may only be the surface. What a creator cannot let go of is dense causal feedback in the making — and that is not the same as controlling every result.
On code-based generative art: when the artist cannot fully control the outcome but remains responsible for it — responsibility without sovereignty — and how Vera Molnár’s “Machine Imaginaire” reframes authorship.
Thoughts on creativity: finding inner light points through accumulated sensitivity, entering flow states, and nurturing fragile inspirations into full works.
A 2010 prose interview fragment. A man tests a mic, then recounts his first outdoor party and first LSD experience in a Hualien valley around 2000.
A 2002 prose fragment. In a city with few green courtyards, a thin man in a long coat naps beside a potted small tree under the afternoon sun, humming and half-speaking sentences no one can finish.