Taiwanese generative artist Aluan Wang posted a lengthy Facebook post on November 13 about a new work he completed with AI, (On the Way to the Art Museum). He wrote that the work accumulated more than 300,000 lines of code over three months, written by thousands of AI work fragments. “I haven’t read a single line of those 300,000-plus lines.”
Halfway through, he began to wonder whether the work could still be considered his. He said the answer he found three months later was simple, and not just for artists: “Among thousands of fragments, only one person was present.”
Wang Hsin-Jen has been creating generative art with code for 15 years. In generative art, the artist does not draw directly but instead writes rules for a program to follow as it draws. In 2021, he became the first Taiwanese artist to be featured on the generative art platform Art Blocks.
(On the Way to the Art Museum): A different street scene with every click
According to Wang Hsin-Jen, he worked on this piece from August to November this year. Each time he clicks, the program draws a different street scene filled with passersby. The AI wrote all the code; his job is to observe, make judgments, say no, and remember what happened the day before.
Street scenes generated by the program in (On the Way to the Art Museum) (Image source: Wang Hsin-Jen’s Facebook page)
A different street scene appears with every click in (On the Way to the Art Museum) (Image source: Wang Hsin-Jen’s Facebook page)
He believes AI is not an assistant that remembers you, but “a copy with no yesterday”: every time you open a new conversation, the same model starts from scratch. You can open ten instances at once and close them when you’re done; none of them accumulate experience. So at the entire work site, he is the only one who remembers yesterday.
The notes written for AI are its entire past
Wang Hsin-Jen says there is a set of notes in his work that every AI must read before starting: what has been done so far, what mistakes were made, and which rules must not be changed. He revised these notes again and again over three months; for the AI, they are its entire yesterday.
Looking back afterward, he also found that when a conversation gets long, the system condenses the earlier exchanges into a summary so the AI can continue working from it. In the end, most of the original process is reduced to a summary written by the AI itself. He admits that he is much the same: he remembers only what is in his diaries, notes, and scripts. The difference is that his summaries were written by someone who actually lived through those days—and he knows he has forgotten things, while the AI reads a summary and assumes it is the whole story.
Not a word from midnight to 5 a.m.: sleep proves he was there
Wang Hsin-Jen arranged everything he said to the AI over three months into 24 hourly blocks. There is not a word from midnight to 5 a.m. because that is when he sleeps. The AI can keep working for hours on tasks it has been given, so the machine continues working while he sleeps. He writes that sleep may sound like “being absent,” but this very blank space proves that the one present is someone who gets tired, sleeps, and remembers yesterday when he wakes up.
Wang Hsin-Jen laid out every word he said to the AI over three months across a 24-hour timeline. There is not a word from midnight to 5 a.m. (Image source: Wang Hsin-Jen’s Facebook page)
He also mentions Taiwanese artist Tehching Hsieh’s 1980 time-clock piece in New York: he punched a time clock every hour on the hour for an entire year. Of the 8,760 punches, he missed 133, mostly because he overslept. Wang Hsin-Jen believes those 133 missed punches are the most human part of the entire work.
Checking can be outsourced; judgment and saying “no” cannot
Wang Hsin-Jen says the AI checks for errors in the people it draws, such as reversed legs or objects falling from their hands. But it can only catch errors that can be measured. Whether a pose looks awkward, the proportions look good, or a particular mistake happens to add character—those are things he has to assess and explain to it.
The next level is deciding what not to do at all. He says that technically he could draw Egypt or the Middle East, but he has never been there, so what he produced would be an encyclopedia illustration, not a sketch from life. So he does not do it. “AI won’t say no. Only you can.”
He emphasizes that this is not a myth about genius: he did not write the code, and what he does is ordinary—show up every day, observe, judge, say no, and remember, then come back the next morning. At the end of the article, he leaves readers with four questions, including what notes they write for AI and when they last said “no.” If you can’t remember, “you may be outsourcing your judgment too.”
This article, “Generative Artist Wang Hsin-Jen: AI Wrote More Than 300,000 Lines of Code, and I Didn’t Read a Single One,” originally appeared at .
【Further Analysis】
【My Further Perspective】
Wang Hsin-Jen’s creative experience reveals the direction of a core shift in value in the AI era: the human role is changing from “executor” to “arbiter” and “keeper of context.” Although AI has tremendous productivity and code-generation capabilities, its limited context window and lack of real-life experience prevent it from forming enduring, subjective value judgments. In an era when the cost of creation and development is falling dramatically, “judgment” (knowing when to say no) and “anchoring a worldview” are becoming the scarcest assets. This logic applies not only to generative art, but also to software development, product design, and financial strategy.
【Market Impact】
* **U.S. equities (bullish):** AI-assisted coding significantly boosts development efficiency, demonstrating the deep real-world industry applications of tools such as AI copilots. This trend is favorable over the long term for major providers of AI development tools and computing infrastructure, such as Microsoft (MSFT), Alphabet (GOOGL), and Nvidia (NVDA). If software development continues to become more accessible, more companies will accelerate their digital transformations, boosting overall productivity across the technology sector.
* **Cryptocurrency and Web3 markets (neutral):** Generative art and on-chain NFT asset markets may become increasingly polarized. AI tools lower the technical barriers to creation, potentially leading to a short-term glut of low-quality NFTs and generative art (bearish). However, the scarcity and collectible value of leading generative artworks with compelling personal narratives, human judgment, and brand premiums—such as high-quality projects on the Art Blocks platform—may become even more apparent (bullish).
* **Taiwanese equities (bullish):** As AI-generated content and computing become commonplace, demand for high-performance computing and servers will continue to grow, providing long-term fundamental support for AI supply-chain stocks such as TSMC (2330) and Quanta (2382).
【What to Watch Next】
Going forward, watch for legal precedents from intellectual property offices in different countries on copyright recognition for “artworks created entirely by AI” and “code written entirely by AI.” Also monitor whether mainstream Web3 generative art platforms such as Art Blocks introduce new review standards and trading rules governing the extent of AI-assisted creation.
【Risk Warning】
If copyright regulations on generative AI assets become more restrictive, valuations of related digital art NFTs could be revised downward. In addition, if the market becomes flooded with low-cost AI-generated imitations, be alert to the risk of liquidity dilution and a bubble forming in the digital asset market.