A few days ago, my journalist friend H sent me a message on Signal, pointing out a website containing information about the Wang Fu Court incident. He said it was very important and was worried the data might be deleted, so he asked me to help back it up. Fortunately, the website structure was simple, so my rudimentary tech skills were just enough to handle it without taking up too much time.
In a Hong Kong that already possesses “democratic elections” (it’s a pity I was very tired this Saturday and had to stay home all day), and in an era where you can ask AI for answers to anything, was H’s worry about a website vanishing into thin air just paranoia?
One Two Three Five Eight Nine Ten
Let’s look at the example of Ta Kung Pao. Yes, that Ta Kung Pao.
Another journalist friend, P, informed me that Ta Kung Pao’s A7 headline story from November 28, titled “Sky-High Fees, Bid-Rigging Collusion, Transfer of Benefits, Shoddy Construction,” had been taken down. I thought they had uncovered some explosive scoop and looked for it with high expectations. Unexpectedly, the sentence most out of sync with the “official main melody” in the entire page was merely: “The tragedy of this fire is a test of Hong Kong’s urban governance capabilities.” Reading from start to finish, the only useful message I got was that the Motherland’s “News Optimization Project” has been upgraded again.
Since nobody reads it anymore anyway, and everyone just asks AI, maybe it doesn’t matter if newspapers are taken down (?) Assuming I knew nothing about the Wang Fu Court fire, let’s see what Deepseek has to say:
Although, like Deepseek, I firmly believe that under the leadership of the Party and the Government, the harmony and stability of the overall social situation will certainly be safeguarded (100 words omitted), firm belief does not mean there are no questions. For example:
- Why did Hong Kong, in its transition “from governance to prosperity,” experience this man-made disaster—a once-in-a-century event with the second-highest casualty count since the 1918 Happy Valley Racecourse fire (and potentially the highest once statistics are finalized), leaving 159 dead and nearly 2,000 families homeless?
- Why are bookstores instantly sued for hosting interest classes, and restaurants fined six thousand dollars at lightning speed for putting tables on the roadside, yet non-compliant construction works continue year after year?
- Why does the ICAC rush to arrest people for encouraging blank votes on social media, yet in a case involving suspected bid-rigging and a construction project worth over 300 million, they only react after a tragedy occurs?
- Why is reporting the news a National Security case, yet a case where many lost their homes and even lives does not involve “National Security”?
Carrying this string of questions, I asked Gemini about the “sequence of events of the Wang Fu Court fire.” The Gemini 3 Flash model responded with 866 words (Chinese). Here is a excerpt, translated:
3. Cause of Fire and Spread (Preliminary Investigation)
- Ignition Point: According to reports, the initial ignition point was the external bamboo scaffolding or protective netting of one of the buildings.
- Key Factors for Rapid Spread:
- Major Renovation Work: Before the fire, Wang Fu Court was undergoing a large-scale renovation costing HKD 330 million. All affected buildings were surrounded by bamboo scaffolding covered with protective netting.
- Flammable Materials: Preliminary investigations indicate that the external protective netting, waterproof canvas, and plastic sheets used by the contractor were suspected of not meeting fire safety standards (insufficient flame retardancy).
- Sealed Windows: To facilitate repairs, many residents’ glass windows were sealed with Styrofoam (polystyrene boards). This material is highly flammable, causing the fire to spread rapidly from the external wall into the interiors.
- Alarm Failure: Reports indicate that the fire alarms in the 8 affected buildings did not sound.
4. Subsequent Investigation and Controversy
- Law Enforcement: Police and the ICAC have arrested a total of 14 people, suspected of manslaughter and corruption respectively, mostly heads of construction companies.
- Project Controversy: Issues regarding the tender for the renovation project, the “black box” operation of the Owners’ Corporation, the role of District Councilors, and the continued use of bamboo scaffolding have all sparked social controversy.
- National Security Department Intervention: The Hong Kong Police National Security Department took away and arrested several people, accusing them of “intending to incite hatred against the government through the fire,” sparking controversy over freedom of speech and law enforcement standards.
Then, I asked ChatGPT and Grok the same question. The responses were similar. To avoid padding the word count (and scamming fees—Mystery Voice: “Pfft, what fees do you have?”), I’ll put the content in a link.
Although the difference is clear, this is not an AI model benchmark. I genuinely think Deepseek is impressive. Even if it lags slightly behind Gemini, OpenAI, and Grok, the disparity in the response above is obviously not technical, but systemic. I don’t blame Deepseek. If Ta Kung Pao‘s headline story gets taken down for not singing praises loudly enough, I—a small potato writing an article—have to speculate with fear and trembling on what constitutes “intending to incite hatred against the government.” If you were Deepseek, even with the world’s top technology, what profound answer could you offer?
So, as long as we have OpenAI, Grok, and Gemini, can we rest easy? Obviously not. Sino-US relations remain tense. Even if the two countries can “make a deal” to resolve trade disputes, AI is a matter of national sovereignty. Large Language Models literally reflect “discourse power” and influence users’ worldviews. A powerful nation would absolutely not want its citizens using models dominated by other countries. They could ban them in the name of national security at any moment. The only reason they haven’t been banned yet is purely that domestic AI hasn’t clearly lost the battle, or other conditions aren’t ripe. Among the three major AIs, Hong Kong people can only access Grok without a VPN, and I am almost certain that Grok will not be spared. If I happen to guess wrong, that’s a good thing—Hong Kong people will at least have Grok. But that would also be a bad thing, as it would mean Grok compromised on certain principles to get the green light.
You might not have realized that the days of the VPN being standard equipment for Hong Kong people have arrived. In fact, by my standards, a VPN has long been a basic human right. If one day Hong Kong people become adept at jumping the firewall, the efficient Legislative Council and the system of “Executive-Legislative Cooperation” will be itching to include VPN use under “dishonest use of a computer” or label authors who educate civic tech as “inciting the public.” I don’t know how Hong Kong people will cope then. I only know that today, while VPNs still work normally and the Consumer Council hasn’t yet taken down the VPN reviews in Choice magazine, if you have money to spend up north but “no money” to subscribe to a VPN, even if your physical space has expanded, your world has actually shrunk.
Conservation in the Now
Simply ensuring access to AI that can speak relatively freely is not enough. To solve the problem at its root, we must start at the source: News Conservation.
It is said that “News is the first rough draft of history.” In the AI era, the logic hasn’t changed, only the writing of history has become a thousand times more centralized, and the process ten thousand times faster. No matter how powerful AI becomes, at this stage and in the foreseeable future, it is merely an organizer with extremely strong literacy skills; it must first have news material as raw ingredients for production.
Focus is concentrated on AI, capital is poured into AI, and revenue is paid to AI. Yet, the raw materials are produced by traditional newspapers, citizen media, and the entire internet population—the latter of which are either struggling to operate or earning zero income. Ironically, AI not only doesn’t give you a share of the revenue, but sometimes it conversely increases your costs. Even my tiny blog incurred an extra USD 33 in bandwidth fees last month due to AI scraping data.
How to solve this dilemma is a huge question I am not qualified to answer. But what is certain is that when citizen media outlets like The Collective HK, InMedia, HK Feature, The Witness, Court News, Hong Kong Free Press, and The Reporter continue to publish openly despite facing various difficulties, we as citizens must support them vigorously. To conserve content, the priority is to let the recorders survive. “Free” doesn’t mean it has no value; it means contributing value for people to access freely, supported by the public contributing what they can through voluntary pricing.
Letting citizen reporting survive is the beginning of content conservation. Subsequent steps include preservation, verification, coordination, curation, and various other issues, in which decentralized publishing plays a key role. If you are a long-time reader, you might have noticed I’ve barely mentioned this topic in the past year. But not talking about it doesn’t mean I’ve forgotten. I am constantly thinking about content conservation. It’s just that experience since 2019 tells me that while frantically backing up content on the verge of disappearing is meaningful, it means being constantly led by the nose and remaining passive.
Take public library book removals as an example. The side holding public power possesses tax resources and can remove large numbers of books at any time without publishing a list. The citizen side guesses the removal list and pays out of pocket to save books, running themselves ragged. I haven’t given up, but while saving what I can, in recent years I have been looking more for ways to embed conservation into the process of publishing and reading, rather than taking it for granted that content will always remain, only scrambling to find help for backups when it’s about to vanish.
Last year’s UBR (Universal Basic Reader) project is one example. In contrast to the Leisure and Cultural Services Department “curating” by removing books worth preserving, UBR takes the initiative, coordinating independent bookstores to select books monthly, which are then purchased and preserved by the public and DHK dao.
Furthermore, 3ook.com and its underlying LikeCoin v3 DeBook protocol are even more radical. Starting from the upload of the text, we apply decentralization mindset to every layer: metadata organization, content storage, distribution, ownership, and curation.
Today’s newspapers and magazines are tomorrow’s cultural heritage. Rather than making a big fuss about talking about content conservation, I hope to merge conservation into decentralized publishing, coordinating recorders, creators, and civil society as a whole to practice conservation in the present moment.
p.s. Last month, I was interviewed by Glasses On (借鏡集). The host, Hinson, is considering writing a book, which gave me the perfect chance to introduce decentralized publishing. The whole interview is nearly two hours long. If you are interested, you might want to save it first and watch or listen to it slowly while commuting. Although YouTube is video, I rarely watch it; I mostly listen. I like sitting on a long-haul bus, wearing traditional wired earphones, listening to Podcast programs I’ve saved up during the week but haven’t had time to watch.
Leave a Reply