Field Notes

My engagement with dance did not precede my work in computer science. It came after years of thinking in terms of formal systems, execution models, and computational control. This ordering matters. Social dance — particularly partner dance with explicit leading and following — exposed the limits of output-driven, predictive coordination in a way that abstract agent models could not. In social dance, stable coordination does not emerge from faster inference or earlier decision-making. It emerges from maintaining a shared temporal envelope — a co-semantic session — where neither party collapses the interaction into unilateral control. ...

This post reflects on a book published in 2015, revisited from the perspective of 2024. The Zero Marginal Cost Society (2015) presents a sweeping vision: Humanity, in order to preserve a clean and sustainable planet, must transition from centralized production to distributed production. We must build highly efficient energy networks, logistics networks, and information networks (the internet), eliminate inefficient labor and unnecessary intermediaries, integrate production and consumption, and embrace open collaboration and shared resources. The goal is to maximize resource utilization, minimize environmental depletion, shift from ownership to access, and prioritize collective ecological well-being over private accumulation. ...

Observation In traditional OEM and ODM processes, suppliers provide a manifest alongside delivered hardware or software. This document functions as evidence that contractual deliverables have been met and is later referenced in acceptance tests or disputes. In AI model delivery, this role is weak or absent. Models are often delivered as opaque binaries, leaving the purchaser unable to verify provenance, training conditions, or behavioral claims. This creates a structural asymmetry: the supplier knows what the model is, while the purchaser bears the operational and legal risk. ...

Field note (retrospective) This document records an early line of reasoning developed around 2018, before the emergence of modern large language models. The intent of this note is to preserve the initial intuitions, metaphors, and concerns that shaped later work, not to assert a current position or policy recommendation. Several assumptions and framings used here have since been revised, formalized, or superseded. Context This essay was originally written from the vantage point of 2018, when breakthroughs in artificial intelligence—such as AlphaGo, advances in speech recognition, and early NLP systems— signaled a structural shift in how humans and machines might collaborate. ...

Abstract As discussions around prompt engineering intensified in early 2023, a popular metaphor emerged: describing users as “AI chanters” who recite incantations to elicit responses from a model. This note argues that the metaphor is not merely inaccurate, but structurally harmful. It mispositions human agency, obscures responsibility, and subtly trains users to relate to AI systems in a subordinate or mystical frame. The Problem with the Metaphor The term chanting implies a ritual act: ...

This field note was written in early 2023, shortly after ChatGPT became widely accessible, at a time when prompt-based interaction was still being actively discovered rather than standardized. Prompt engineering has quickly become a widely discussed topic. Alongside collections of reproducible prompts—tested, shared, and refined by communities—there has also been an effort to provide theoretical guidance for users without a computer science background. It has become increasingly clear that prompting is not a peripheral skill, but a central operational capability in human–AI interaction. ...

When language becomes the primary operational interface, human value no longer resides in parameter tuning, but in narration, selection, and endorsement. This observation predates large language models. I first arrived at it around 2017, at a time when LLMs did not yet exist in their current form. What was already evident, however, was the rise of systems such as AlphaGo and Generative Adversarial Networks (GANs). These systems demonstrated that structure, evaluation, and meaning could be negotiated without explicit symbolic programming. In this sense, images were already functioning as a form of language— not descriptive language, but operational language: a medium through which intent, constraints, and outcomes could be expressed, tested, and refined. ...

Abstract This note records an early operational observation (December 2022) on how large language models alter human work not by replacing labor, but by shifting humans from execution to delegation, constraint setting, and judgment. At the time of writing, the language of “agents,” “AI workforce,” or “task orchestration” was not yet mainstream. The observations below precede those terms, but describe the same structural shift. A Common Misreading When ChatGPT became publicly available, most early reactions fell into two camps: ...

This piece was originally published on my early personal blog in 2008. It reflects an early intuition that programming is fundamentally a cognitive and symbolic activity, and that human value in computing should not be tightly bound to any specific physical interface. Long before voice interfaces or large language models existed, this question already pointed toward a more general idea: interfaces are replaceable; cognition is not. （中文）口足程式設計師 如果沒有手，你不知道什麼是擁抱的感覺。 如果沒有手，你不知道生活中許多理所當然的動作原來都這麼麻煩。 如果沒有手，你將是什麼樣的人？ ...

Field note, written in 2005 during my university years, reflecting on online identity, disposability, and responsibility in early BBS culture. 中文原文（2005） 語氣、詞彙、以及藏匿在句子中的思想， 在 1 與 0 的空間中交織出一個網路人格。 於是我確信我真的從一篇篇的觀想， 漸漸描繪出一個概略的輪廓，認識了你。 只是你永遠是那樣的模糊不清晰， 你終究只是個 ID。 而我也靠著一次又一次的貼文， 雕塑出一個網路人格，讓別人認識部分的我， 這也終究只是個 ID。 或許有天醒來， 突然拋開上頭承載的人格、信譽、人際關係， 重新註冊一個 ID，重新開始。 那，這算不算一個生命的死亡？ 是不是一個 ID 誕生得太容易， 所以學會了說話不負責， 久而久之也將這個惡習， 從虛擬中帶進現實？ 我們處於一個尷尬的年代， 徘徊在虛擬與真實之間。 English Translation Tone, vocabulary, and the thoughts hidden within sentences interweave in the space between ones and zeros, forming an online persona. ...