Notes

Based on the nanoMuse v1.0.9 source (main @ 067384b), this post dissects how an open-source personal agent implements proactive intelligence: a scheduler loop waking every 30 seconds handles all due checks; due items are handed to the model through an idempotent SQLite ledger; the model's output goes through the [quiet] prompt protocol that decides whether to interrupt the user; dangerous actions are stopped by the Sentinel gate. The most counterintuitive finding: a failed model run still counts as 'this occurrence happened' — only a process crash gets replayed. Conclusion: the reliability of proactive intelligence comes from a deterministic engineering shell, not from the model.

This essay examines what remains uniquely human in the AI era. As AI drives the cost of execution toward zero, the core question shifts from 'can you do it' to 'what is worth doing, what counts as good, and what should come first'. The author argues three things cannot be outsourced: taste, which judges what deserves to exist; curiosity, which asks good questions in an era of surplus answers; and agency, which moves first and owns the consequences — all resting on the courage to own your choices. Tools no longer differentiate people, yet knowing their boundaries still matters; the two kinds of truly unoutsourceable cognition are defining problems and verifying outputs. How you use AI reshapes cognition itself: replacing your thinking with it atrophies cognition, while using it to pressure your thinking compounds it. The essay closes with a motto: ask with curiosity, choose with taste, act with agency; leave the rest to the machine.

jabinwan's personal site goes live: a bilingual (Chinese/English) personal site deeply customized from the Refined-X (Astro + Starlight) template. The static site runs on Cloudflare Pages; a lightweight visit counter is served by a Cloudflare Worker.