I had done a little homework before the trip and knew the library was not far from Tagajo Station. What I didn’t realise until I arrived was that the two were basically next door to each other. For someone from Hong Kong who assumes anything attached to a train station must be a shopping mall, that alone was a surprise.

If, like me, you still associate libraries with solemnity, seriousness and a certain institutional dullness, Tagajo City Library will completely change your mind. The first thing that caught my eye was its architecture: clean lines, restrained materials and a surprisingly sophisticated feel.

Opened in 2016, the library is the centrepiece of the redevelopment around Tagajo Station. Following the 2011 Great East Japan Earthquake, it came to be seen as a symbol of the area’s recovery.

Libraries generally have strict no-food-and-drink rules, and the more traditional ones may not even allow you to talk. Tagajo City Library, by contrast, brings together a bookstore, café, co-working lounge, restaurant and shops, openly encouraging people to eat, chat and spend money. If you insisted on calling the place a small shopping mall next to the station, you wouldn’t exactly be wrong.

Before you even see a bookshelf after walking through the entrance, you are greeted by lifestyle brands and Starbucks. What kind of library is this?

And the gauntlet doesn’t end there. After the shops and café, you still have to pass through the bookstore before reaching the actual library. Calling it three stages isn’t an exaggeration. With Japanese design and displays everywhere, plus all the latest manga, anyone with even slightly weak resistance to consumer temptation might end up making an “accidental purchase” before ever reaching the library collection.

Tagajo City Library is three storeys tall. The higher you go, the more serious it feels — and the more it starts to resemble the library I had in mind.

The new school term had just begun, and I visited on a Sunday, so for whatever reason, there seemed to be a lot of people doing homework and studying.

Perhaps it’s something to do with Japanese culture, but even with so many people around, the place remained remarkably quiet without ever feeling lifeless. Instead, there was a sense of energy that felt very different from my usual experience of visiting libraries in Hong Kong.

The unconventional approach at Tagajo City Library is not simply a matter of creative interior design. It stems from Japan’s 2003 revision of the Local Autonomy Act, which introduced the “designated administrator system”, allowing private companies to operate public facilities such as sports centres and libraries. In broad terms, it is a form of public-private partnership.

Japan has around 3,300 public libraries, of which roughly 700 are operated by designated administrators. The model can broadly be divided into two types. The first consists of professional library-service providers responsible for cataloguing, circulation systems and the dispatch of qualified librarians — roughly equivalent to outsourcing in Hong Kong. The work itself remains much the same, but the staff are employees of private companies rather than civil servants. The dominant player in this field is the Library Distribution Centre (TRC), which accounts for more than 70 per cent of the market.

The second type is the kind represented by Tagajo City Library: a lifestyle-oriented concept that takes a completely different approach to running a library. Its best-known operator is CCC (Culture Convenience Club), the parent company of Tsutaya Books, which is more familiar to readers in Taiwan.

Besides Tagajo, CCC also operates libraries in Takeo, Saga Prefecture; Ebina, Kanagawa Prefecture; Takahashi, Okayama Prefecture; Shunan, Yamaguchi Prefecture; Uki, Kumamoto Prefecture; and Wakayama City. Most have a Tsutaya Books and Starbucks on site, with an emphasis on aesthetics and spatial design. They also extend opening hours into the evening and stay open year-round. Several, like Tagajo City Library, are located next to railway stations, making them part of station-area redevelopment and local revitalisation projects.

But does bringing in private companies to make libraries more accessible, fashionable and integrated into everyday life really come with no downside? Even as a passing tourist with only the vaguest knowledge of Japan, I knew it couldn’t possibly be that simple.

More visitors, greater efficiency and better cost control are all positive developments. But library collections are supposed to maintain a balance — and arguably should even lean towards serious, niche and commercially unviable works that nevertheless have cultural value. Under a commercial management mindset, can libraries still uphold their role in preserving culture and safeguarding history? That remains a big question.

And then there is the fine line between public-private partnership and cosy government-business relationships. Tagajo City Council has previously seen councillors question why the city government appointed CCC without an open tender. The education superintendent at the time responded that the decision was made as part of the station-area redevelopment and that the procedure was therefore appropriate.

By comparison, public libraries in Hong Kong and Taiwan are still largely operated directly by the government, apart from services such as cleaning and security that may be outsourced. The Kaohsiung Main Public Library, which has moved towards an administrative-corporation model, is a notable exception. My visit to Tagajo City Library may have started as a bit of movie-location sightseeing, but it certainly opened my eyes to new possibilities for imagining what a library could be.

I’m not naive enough to think that commercialisation is automatically a good thing. But I do believe that if the private sector can participate in building and operating libraries in some form, it could bring a breath of fresh air to the way we think about them.

p.s. From Bring Her down to a Portable Size and Postcard from Sendai to this week’s newsletter, my Japan trilogy has come to an end for now. Next week, both my body and soul will return to Hong Kong, where I’ll be talking about out-of-print books. But before that, I still have one more dose of Japan fever to get through: I’m heading to the cinema to see Hirokazu Kore-eda’s version of Look Back. Can’t wait.

We mint POAPs so people can own their stories

POAP serves as a personal proof of attendance for events—or in plain terms, a “digital souvenir badge,” or colloquially, “Web3 check-in.”

Short for Proof Of Attendance Protocol, the immediate whiff of technical nuance gives it away: smart readers will guess it originated right from the mind of a tech geek.

In February 2019, during the ETHDenver hackathon, Patricio Worthalter from Buenos Aires first proposed recording attendance credentials on the blockchain. He extended the concepts of Proof of Work and Proof of Stake from computation and capital to physical and digital “presence,” coining the term Proof of Attendance Protocol.

Together with a few teammates, Patricio hacked together a POAP prototype within 72 hours of the hackathon and deployed the initial smart contract to the Ethereum mainnet. Utilizing what was on hand, they had several hundred attendees on-site to scan QR codes and collect the very first batch of POAPs in history—the “ETHDenver 2019 POAP.” It received unanimous acclaim and swiftly became the standard check-in fixture for Web3 community events.

To ensure a seamless minting experience, POAP utilized meta-transactions to cover gas fees on behalf of users. These early costs were absorbed personally by Patricio, who treated POAP as a public good. However, during 2020’s DeFi Summer, transaction fees on the Ethereum mainnet skyrocketed to several dollars a pop—especially for ERC-721 interactions requiring more computational overhead. It is estimated that Patricio personally poured two to three million dollars into gas fees and infrastructure costs for the community. Consequently, in late 2020, POAP migrated to the Ethereum sidechain xDai (later renamed Gnosis Chain), drastically lowering gas costs while absorbing them through xDai subsidies.

Over two years, from event check-ins and token-gated community access to serving as criteria for early-adopter airdrops across major projects, POAP gradually evolved into an “on-chain resume” for Web3 users.

Conceived in 2019 and established in 2021

However, as POAP transitioned from a niche passion project to a core piece of infrastructure serving millions worldwide, the weight of reality set in. The team realized that running an ecosystem of this scale was vastly more expensive than envisioned.

The primary challenge lay in Sybil attacks and rampant airdrop farming. Because collecting a POAP was entirely free, projects issuing airdrops to holders naturally attracted vast swarms of bots and farming studios. Even when a small gathering of dozens of people issued a POAP, thousands of addresses on-chain would try to rush in and claim it.

To preserve the authenticity of attendance and participation credentials, POAP built a manual review team and introduced stringent anti-bot measures—including dynamic Kiosk QR codes and CAPTCHA verifications—which drove up operational and server costs immensely.

With the NFT boom of 2021, even though POAP was never meant for speculation, issuance and user numbers exploded exponentially. The project could no longer rely solely on Patricio’s personal pocket and volunteer enthusiasm. In late 2021, POAP Inc. was officially incorporated, led by co-founder and COO Isabel to spearhead commercialization, announcing a $10 million seed round in January 2022.

Following that, POAP began collaborating with enterprise clients, customizing on-chain check-ins for traditional giants like Adidas, Porsche, Amex, and Goldman Sachs, while introducing pay-per-mint quotas and SAML enterprise-grade authentication. What was originally a free distribution model and simple collection process became increasingly complex.

Unsurprisingly, this pivot met immense resistance. On one hand, a spoiled Web3 community (myself included, likely) had long taken for granted that POAP should be free and unrestricted; as partial paywalls were put up and review thresholds raised, users departed rapidly. On the other hand, enterprise solutions suffered from long sales cycles and one-off payments, failing to build steady recurring revenue. Compounded by fading Web3 narratives, POAP’s operations grew increasingly difficult—and the 2026 bear market was merely the final straw that broke the camel’s back.

Hard to build a sustainable company without cannibalizing the ethos

To talk about POAP, one cannot overlook the geek ethos of its guiding light, Patricio. Yet, it was precisely this ethos that laid the groundwork for POAP’s farewell.

As a seasoned early Ethereum advocate from Argentina, Patricio is a pure crypto philosopher. His original intent for founding POAP was to record genuine human experiences and interpersonal connections on-chain. He firmly insisted on a non-financialized model, repeatedly publicly criticizing airdrop hunters who treated POAP as a speculative tool, and consistently refusing to issue a governance token in an era when almost every Web3 project went for ICO to raise fund. Had they issued a token, the team might still have ample runway to keep exploring the space between reality and ideals.

In her farewell tweet, Isabel reflected: “crypto’s funding cycles and distribution dynamics made it hard to build a sustainable company without cannibalizing the ethos that made POAP mean something.” Though understated, reading between the lines conveys a clear message: “Because we refused to compromise the ideal POAP stood for, we could not sustain the company.” To loosely translate: they preferred to end this five-year idealist experiment rather than trim their sails to fit reality.

Patricio’s principles earned POAP immense respect across the Web3 industry, but they also prevented POAP from turning into a sustainable commercial service. This is the tragic fate of Web3 innovation today: when ideals cannot be monetized and public goods lack sustainable funding mechanisms, even the most beautiful products eventually get swallowed by reality.

Even as the official servers and apps shut down, the 7+ million POAPs already minted remain forever on Gnosis Chain—a permanent testament to our generation’s exploration of Web3, proving we once truly crossed paths and dreamed together.

p.s. In last week’s poll asking which profession people are warned to watch out for most, bookstore clerks took first place with 63%, followed by Web3 practitioners at 31%, leaving triads (3%), scaffolding workers (2%), and soldiers (2%) far behind. I have a friend who happens to be both a Web3 practitioner and a bookstore clerk—what on earth is he supposed to do?

Ever since Bitcoin reached a new high last year and began its step-by-step decline, we have entered an undeniable bear market. Bitcoin hit a low of $58,410, a 52% drop from last year’s all-time high; Ethereum hit a low of $1,546, falling 65% from its 2025 peak. That said, compared to Bitcoin dropping 77% in 2022 from its 2021 peak, and 84% in 2018 from its 2017 peak, this is currently the smallest peak-to-trough decline among the major bear markets since Bitcoin’s birth in 2009—suggesting the market is entering a phase of maturity.

However, in terms of the ecosystem, 2026 is undoubtedly the most brutal, nerve-racking bear market in history. According to statistics from RootData, in the first seven months alone, over a hundred projects have announced shutdowns, bankruptcies, or indefinite suspensions. And these are only the ones that can be tallied; countless others have quietly vanished without a trace. Price fluctuations are often just investors overreacting, but the departure of industry practitioners reflects the reality of the market.

Exchanges Flash-Exit, Star Projects Say Collective Goodbyes

Let’s start with centralized exchanges. Founded in 2018 and headquartered in Singapore, AscendEX shut down on July 1 after failing to secure licenses from regulatory bodies like the Monetary Authority of Singapore (MAS) or Europe’s MiCA. Even more surprising than AscendEX was BitMEX—the pioneer of perpetual swaps founded in 2014—which announced on July 23 that it would be winding down operations by September 23. A window of just two months from announcement to closure is extraordinarily short for a major exchange. A mere three days after that, BitMart, known for its rapid token listings and vast array of altcoins and memecoins, announced on July 26 that it was halting deposits immediately and would fully shut down by January 31 next year.

If it were only exchanges closing down, the situation wouldn’t be quite as alarming. However, user-facing application projects are suffering an even worse wave of shutdowns. Among them, what I regret most is Tally, the DAO governance platform that served Arbitrum, Optimism, Uniswap, Aave, Gitcoin, ENS, and was almost chosen by LikeCoin during its v3 upgrade. Another favorite of mine, the “Web3 check-in” service POAP, while not completely shutting down, entered maintenance mode in March, ceasing new feature development and disabling the issuer distribution interface. Furthermore, decentralized storage provider Storj filed for Chapter 11 bankruptcy protection just the other day—on the very same day BitMart announced its shutdown—in an attempt to restructure its debt and keep operating.

Tally, POAP, and Storj are all services built on solid tech that genuinely met user needs, yet they still failed to evolve into sustainable businesses before their runway ran out. Products still in the exploratory phase fared even worse: Goldfinch (credit), Slingshot (trading), Nifty Gateway (NFT marketplace), Fantasy Top (card game), Dmail (decentralized email), and Yupp (AI model feedback)—all star projects that had raised massive VC funding—have all shut down one after another this year.

Beneath a Collapsing Nest, How Can Any Egg Remain Intact?

With applications on the decline, data analysis and tracking platforms naturally haven’t fared any better. This year’s closures include Parsec and the even more famous Zapper, which once boasted over two million monthly active users.

On the wallet front, at least three major wallets have shut down in 2026: Leap Wallet, Family Wallet, and Ctrl Wallet. Among them, Leap was a Cosmos ecosystem wallet that served LikeCoin v2 and is likely familiar to readers; its founder, whom I once met briefly, is a pragmatic entrepreneur. Unlike standard Web3 applications, wallets manage users’ private keys. If developers aren’t vigilant, a slight delay in patching a vulnerability can lead to disaster. If readers have ever used any of the above wallets, I strongly recommend discontinuing their use immediately and withdrawing your assets if necessary.

Finally, there are the blockchains themselves. The developer company behind Movement, the first Move-language L2 on Ethereum—MVMT Labs—filed for bankruptcy on July 15. Although assets related to Movement have been transferred, whether it can continue operating remains an open question. As for other Ethereum L2s, which were already facing existential crises due to the mainnet becoming increasingly cheap, the bear market has made survival even tougher. The first to shut down was Polynomial Chain, and more are likely to follow. On the other side, the Cosmos ecosystem, built on an app-chain design, has become a disaster zone for L1s. Beyond the confirmed shutdowns of MilkyWay (liquid staking) and UX Chain (formerly Umee), many others are effectively dead—they just haven’t made official announcements or set up exit mechanisms.

The Responsibility of Exiting, and Next Steps for Players on the Field

It’s one thing not to list them, but seeing them all laid out is terrifying—it’s like being able to digest one slice of cake a day, but suddenly having to swallow a hundred at once. No appetite is big enough for that.

What’s that? You haven’t seen news about Web3 projects shutting down? Perhaps it’s because so many projects are closing that it’s no longer considered news, or maybe the Web3 media outlets you used to follow have stopped updating or pivoted elsewhere. Speaking of media, perhaps because hosting old content costs very little, Chinese and English Web3 media rarely declare an official end. In reality, however, an exceptional number have quietly ceased updates or shifted focus to writing about AI. I’m certainly not trash-talking here—everyone has bills to pay, and stopping updates or pivoting makes complete sense from a business standpoint. It’s totally understandable.

Fortunately, even in such tough times, fellow writers like Max (Little Crow) and Astro Hsu (BlockTrend) continue to produce great articles week after week. Keeping Web3 at their core, they hold down the fort, ensuring that as I absorb industry news during the day and rush to finish my newsletter in the dead of night, I don’t feel alone.

The whole article might sound full of gloom and doom, but the survival of the fittest is simply the law of the market. I am not overly pessimistic about the current state of affairs. On the contrary, I respect projects that formally announce their shutdown, properly arrange asset withdrawals or transitions to other services, and see things through to the end. That is far more responsible than turning into a zombie project that gives no explanation, handles nothing, and acts as if it no longer exists simply by never mentioning it again.

Unlike Max and BlockTrend, I am not only a commentator but also an active player on the field. Readers naturally ask me where LikeCoin is headed in the face of such industry hardship. I won’t dodge this sharp question, nor do I want to gloss over it at the end of this article; I will address it in a separate post later.

With no end in sight for this crypto bear market, and capital, talent, users, and attention all focused on AI, the Web3 industry seems to have lost its collective voice—no matter what vision is articulated, there are no listeners. In the days ahead, projects at every level will have no choice but to keep their heads down, build good products, and actively acquire users. They will either erupt out of the silence or perish within it.

p.s. At 2:00 AM, as I was writing this draft, news broke in the bookstore group chat that Tin Yuen Bookstore was closing down. Rooted in a Mong Kok tenement building for fifty years, quietly nourishing generation after generation of readers, Tin Yuen Bookstore—which could easily be considered intangible cultural heritage—is saying goodbye to Hong Kong due to murky red lines and shifting enforcement standards. It isn’t just Web3 projects that will either erupt out of the silence or perish within it.

On June 24, 2021, Hong Kong’s Apple Daily was shut down, and my column #decentralizehk came to an end along with it. Unwilling to simply stop writing, I launched DHK Newsletter #0 the very next day, June 25. Since then, I’ve written one piece every week without a single break. In other words, this issue, #261, marks DHK Newsletter’s fifth anniversary.

Following past tradition, I’ll close with an annual review.

Previous reviews: Year 1 · Year 2 · Year 3 · Year 4

1. I can’t quite tell whether these five years have passed quickly or slowly. Two contradictory feelings circle me at once — one says five years blinked by in an instant, the other says I’ve finally made it through five years. The objective fact is: although I once wrote a newspaper column for nearly a decade, five straight years of weekly output is, for me personally, the most sustained record I’ve ever kept.

2. Since I choose my own topics, I already knew that this year, among the “three freedoms”, I leaned more heavily toward “freedom of information.” But it wasn’t until I just tallied everything up that I realized how lopsided the distribution actually was — articles related to “freedom of information” made up half of the total, followed by “financial freedom” at 27%. Pieces on “democracy” were fewer than in previous years, only 6 articles, while “Diary of a Fool” stayed at around 10% as usual, also 6 articles.

3. This year there were especially many articles on “freedom of information” — partly because of the LikeCoin v3 update, around which I wrote ten pieces, and partly because over the past year authoritarian power has grown ever more brazen: bookstores raided, the press suppressed, with related incidents occurring more frequently than ever.

4. As in past reviews, below I list all the year’s articles, with 19 specially recommended pieces marked out. That might seem like a lot, but they’re all important topics, so I won’t limit myself to a fixed number. I’d suggest saving this post and coming back to reread it when you have time — branching outward from it, you can connect to all the articles from these five years.

5. For six months of this past year, alongside writing the weekly Post, I was also preparing my new book The Pursuit of Freedom. Though I call myself a “full-time writer,” I write extremely slowly, and on top of that there’s 3ook.com and Just Books to run part-time. With everything piling up, the newsletter sometimes felt like a real struggle to write — but I still managed to publish on time every week, and never once let AI ghostwrite for me.

6. I believe the topics I’ve chosen matter, and I’ve done my best to “tell the web3 story well.” And yet, over this past year, paid subscribers haven’t grown at all — worse, the open rate has kept declining, dropping from the low 30s to the low 20s percent within just a year, a drop of over 20 percentage points. Honestly, I have almost no idea what to do about this. But “knowing shame is close to courage” — so on the occasion of this fifth anniversary, I want to let everyone still reading know about this trend, so we can face it together.

Freedom of Information (25)

• 💈 Will Taiwan and Hong Kong Disappear in the Age of AI? 💈

• An Old Salt’s Upgrade Diary: Is the AI in WordPress 7.0 Any Use?

• 💈 Hong Kong Sends Undercover Agents Into a Bookstore 💈

• 💈 Dear Algo, What Is an Algorithm? 💈

• 💈 Travelling Back to the Qin Dynasty to Maintain Stability 💈

• 3ook.com – Give books a voice

• 💈 One Punch to Wake the Readers: Independent Bookstore Facing One Lawsuit After Another 💈

• 💈 A Slave to Freedom, or A Free Slave? On Jimmy Lai of Apple Daily💈

• LikeCoin v3 Upgrade Accomplished: Heading Toward the 8th Minting Anniversary

• Just Books — The Bookstore That Sells eBooks

• Paper Books Sold Out, E-books Take Over: The Dignity and Meaning of Out of Print

• Silent Night, Upgrade Night. My Fear Of, Your Missing Out.

• Decentralized Publishing: The YouTube of Printed Media

• Stocks have IPOs, Tokens have ICOs, Books Need IBOs

• 💈 Reflections on the Hong Kong Deadly Fire: Today’s Newspapers and Magazines, Tomorrow’s Cultural Heritage 💈

• Thanksgiving to the Preservers of Culture: Airdropping ETH and LIKE

• Acting as a Human ATM: The Price Scare on LikeCoin v3 Launch Day and My Rookie Mistake

• A Big Question for Creatives: Should You Publish in Print or Digital?

• Upgrade to LikeCoin v3 and Share a 10,000,000 LIKE Incentive Pool (Full Tutorial Included)

• LikeCoin v3 Upgrade Countdown: 5 Days Left. See You at the Launch Party on November 3rd.

• LikeCoin v3 Whitepaper Officially Submitted for Community Vote

• Imperfectly Completed: New Book “The Dao of Games”

• Embedded Wallets: The Account is the Wallet, Bringing Web3 to 5 Billion Netizens

• A Soft Resistance Against iPhone’s Dominance: I Switched to a Black-and-White Phone

• LikeCoin v3 Whitepaper in Plain English

Financial Freedom (14)

• The One Thing Bitcoin Holders Need to Do in the Face of the Quantum Computing Threat

• Privacy: A Feature for the Virtuous, or a Bug for the Wicked?

• Please, Just Take the Money: My 8-Year Struggle to Give Away Web3 Red Packets

• Celebrating the Year of the Horse: 188 USDC Red Packets Giveaway

• 💈 The Underlying Logic of Queuing Gangs: Proof of Work💈

• From Gaming Clans to Web3 Guilds: Venturing into the Base Chain World

• A Guide for Base “Immigrants”: Start Your Base Chain Life in Three Steps

• Both sides now: Coinbase = COIN + Base

• 💈 The Eight Great Misconceptions About Money: Cryptocurrency Edition 💈

• 💈 Eight Misconceptions About Money: The Fiat Currency Edition 💈

• 💈 One Last Gift, Part 3: Avoiding Single Point of Failure with Social Recovery 💈

• 💈 One Last Gift, Part 2: Summoning Shenron and Finding the Dragon Ball Guardians 💈

• 💈 One Last Gift: A Decentralized Plan for Your Cryptocurrency Inheritance 💈

• Five Years of “The 5-Minute Sovereign Account”

Democracy (6)

• Instagram Removing End-to-End Encryption: Not a Feature Adjustment, but a Precision Harvest

• 💈 The price of privacy? HK$100k and 1 year in prison. 💈

• 💈 Legitimacy Beyond Legality: On DAOs and Association, from the NASU Referendum 💈

• Substance Over Form

• 💈 A Decade for Ether, A Century for Humanity: Ethereum’s 10th Anniversary, Part II 💈

• 💈 A Decade for Ether, A Century for Humanity: Ethereum’s 10th Anniversary 💈

Diary (6)

• 💈 Imperfectionism: Ship First, Let the World Help You Improve 💈

• Bitcoin Pizza Day | eMPF | Sandy Lam’s Resonance

• The Premier Hub for Mega Events, The Worst Hub for Ticketing

• Shino Lin “Tempting Heart Coordinates” Concert

• 2026 Main Quest: Books and Push-ups

• 2025 Report Card

p.s. Looking back, freedom turns out to be right there, where the lights grow dim.

How Claude Was Forged

Anthropic has always occupied the moral high ground, emphasising the safe deployment of AI and registering as a Public Benefit Corporation (PBC) rather than an ordinary for-profit C Corporation. The company, ever insistent on the proper use of AI, frequently accuses Chinese companies of distilling Claude — acquiring American research and development on the cheap. But if Chinese models were distilled from Claude, how did Anthropic train Claude in the first place? A copyright lawsuit offers some clues.

In August 2024, three American authors sued Anthropic for using vast quantities of both legitimately purchased and pirated books to train its large language model, Claude. Court proceedings revealed that in February 2024, Anthropic poached Tom Turvey, the partnerships lead on Google’s book-scanning project, and bought up enormous numbers of mostly second-hand print books, scanned them, extracted the text with optical character recognition (OCR), and destroyed the originals. At the same time, Anthropic downloaded more than seven million pirated books from shadow libraries such as Library Genesis.

In June 2025, a federal judge ruled that scanning and destroying legitimately purchased print books amounted to format shifting, and that digesting huge volumes of text to learn the structure of language — conceptual machine learning — was highly transformative. Both, the court held, were fair use permitted under the law.

As for the pirated books: even if the training itself was lawful, acquiring and retaining illegal data sources was not. In late 2025 the two sides reached a settlement, with the authors receiving more than US$1.5 billion. The case was closed — and several million legitimate print books, plus seven-million-odd pirated e-books, became the nourishment of today’s leading large language models.

Ordinary people like me have no real sense of what US$1.5 billion means. For reference: in September 2025, shortly before the settlement, Anthropic had just raised US$13 billion. And according to estimates reported by the Wall Street Journal, Anthropic’s revenue in the first quarter of 2026 was US$4.8 billion, with the second quarter expected to reach US$10.9 billion — more than doubling.

The More Law-Abiding, the More Absent

On the subject of pirated books, let me share a personal experience. A few years ago, to free up space and simplify my life, I decided to part with a batch of print books, while trying to keep digital copies of each. Working through my own book list, I searched online for pirated versions one by one. The sources were much the same as those Anthropic used — shadow libraries such as Library Genesis. The result: most English-language and mainland Chinese books could be found, but pirated copies of Hong Kong and Taiwan books were few and far between.

From this little story, some will read pride: copyright awareness in Taiwan and Hong Kong far exceeds that on the mainland. Others will read “underlying logic”: the Taiwan and Hong Kong markets are simply too small for piracy to reach commercial scale. Either way, when legitimate copies dominate and piracy never takes hold, rights holders win.

This essay’s concern, however, lies on another level: the fact that shadow libraries like Library Genesis carry hardly any Hong Kong or Taiwan editions means that, among the material Claude learned from, Chinese-language books come overwhelmingly from the mainland — missing the local perspectives of Hong Kong and Taiwan. In a market this fiercely competitive and this transparent, the other leading large language models are very likely in a similar position.

The more law-abiding, the more refined, the more niche a cultural sphere is, the more easily it ends up structurally absent from AI’s collective unconscious — a deeply ironic yet entirely real cultural phenomenon. To say that Taiwan and Hong Kong may disappear in the age of AI is not scaremongering but genuine anxiety: fewer and fewer people pick up books, more and more rely on AI for their information, and yet Taiwan and Hong Kong editions are all but absent from AI’s training material. How can that not be cause for worry?

Free as in Freedom

Of course, the training material for large language models goes beyond books: public web pages, forums, code and more. PTT, Dcard, Bahamut, LIHKG, HKGolden, open Substack newsletters and blogs — all of it is fodder for AI. The absence of books — the most information-dense of long-form texts, cultural expression distilled over years — may well affect the deeper frameworks of narrative and values. Yet mainstream AI can still read and write traditional Chinese, including Cantonese, idioms, even memes, and can keep its finger on the pulse of the times.

In other words, Taiwan and Hong Kong still exist, for now, in AI’s worldview — and the key is the open internet. Civic media outlets that bear heavy operating costs yet insist on keeping all their content open — Hong Kong’s The Collective (集誌社), InMedia (獨立媒體), Court News (庭刊), The Witness (法庭線), HK Feature (誌) and Hong Kong Free Press, and Taiwan’s The Reporter (報導者) — not only provide open content directly to readers, but also serve as nourishment for machine learning, supplying the major AIs with Hong Kong and Taiwan’s local perspectives, and thereby serving the public indirectly.

The point of open information is freedom, not free of charge. Under the shadow of authoritarianism, every one of these civic media outlets struggles to keep going; their daily reporting on society can never be taken for granted. To ignore the importance of paid support is a society’s slow suicide. One day, not only will the civic media fold one by one — Taiwan and Hong Kong themselves will disappear in the age of AI.

Support The Collective’s 1:1 Matching Campaign

What matters is action. Before the end of this month, buy the e-book The Collective: Our Records on the Ground (《集誌——我們在地記錄》, HKD 140) at 3ook.com, and add any extra amount of support you wish. I will personally match, 1:1, both the book price and any additional amount, paid directly to The Collective, up to a cap of HKD 120,000.

Example: if 100 people buy The Collective: Our Records on the Ground at 3ook.com before the end of the month, each adding HKD 1,000 in support → I will match 140 × 100 + 1,000 × 100 = HKD 114,000.

購買《集誌——我們在地記錄》

p.s. Not wanting to steal the spotlight from the civic media in the main text, let me say this quietly down here: DHK Newsletter aims to provide web3 civic education and to supplement discourse beyond the official narrative. It, too, keeps all its articles fully open — and needs your support.

Like everyone else, I love perfection. But if I had to file myself under a label, I’m probably an “imperfectionist.”

Perfection Is a Direction, the Pursuit a Journey

Owning up to being imperfect — pretty shameless of me, you’ll say. It isn’t. Hear me out.

Take writing. If I claimed every issue of the DHK Post were perfect, that would be the more shameless claim by far. The truth is, while I’ve always published on time, every week, without fail, some pieces I’m not too happy with and some I’m quite pleased with — and even the one I’m proudest of isn’t perfect; you’ll probably still turn up a typo.

But that’s only my own taste, and the paradox is that it’s often out of sync with how readers react. I rather liked last month’s “Dear Algo: What Is an Algorithm?”, yet the response was lukewarm; whereas “Hong Kong: Capital of Mega-Events, Third-Rate Ticketing” — a piece I didn’t even like, finding it too much navel-gazing — somehow became, of all things, the most “CLS”-ed piece I’ve ever written.

Judging a piece of writing takes in many dimensions. To be perfect is, of course, to do the best on every one of them — but those dimensions are forever in conflict, beyond the author’s control. Which is to say: an author can never wait until something is perfect to publish; he can only do his best. And anyway, missing your publishing cycle in pursuit of perfection is, in itself, imperfect.

Imperfectionism is the realization that “perfection” is a state you can only pursue, never reach — like “infinity” in mathematics: go ahead, write down an enormous number, but any number you can actually write down is finite. Where perfectionism treats “perfection” as the destination, imperfectionism cares more about the process of pursuing it.

Imperfectionism is not a refusal to pursue perfection — still less a pursuit of imperfection. It is striving toward perfection inside a thicket of constraints: limited time, limited resources, conditions beyond your control, invisible red lines.

With all that groundwork laid, you can see why I so often hand in homework riddled with flaws. In my eyes, ship-then-improve beats trying to be “perfect” before you ship; and more importantly, only this way can the audience take part and collaborate, finding consensus along the way. To me, the process of public participation matters far more than the result of going it alone. As Audrey Tang often quotes, from the Canadian poet-songwriter Leonard Cohen: the crack is how the light gets in.

Forget your perfect offering.
There is a crack in everything.
That’s how the light gets in.

— “Anthem,” Leonard Cohen

Letting AI Learn from Its Mistakes

Imperfectionism isn’t the reason my work falls short (though it does make a handy excuse now and then) — it’s my worldview, and it shows up in every corner of life.

On June 16 last year, “3ook.com,” the “third-generation bookstore,” launched. Plenty of people found it pretty crude: a thin catalog, few features, no companion mobile app — all sorts of shortcomings, in short. Even so, we kept our rhythm and shipped on schedule.

Fifty weeks — that is, 25 development cycles — later, 3ook.com’s catalog has grown enormously; we’ve added personal voice artists and many other features, and released apps for Android, iPhone, iPad, even macOS. We’re still a hundred thousand miles from perfect, but the point is that we’ve taken many steps in perfection’s direction.

A perfectionist would argue we should have waited until the catalog was large enough, the features complete, every platform supported, before launching — the way Apple, that supposed paragon of enterprise, surely polishes a product to perfection before release. But don’t forget: we’re not Apple. And besides, the iPhone couldn’t even do 3G when it launched in 2007; iOS 26 was bumped to 26.1 not long after release; the Vision Pro still comes with an external battery to this day. All of which goes to show that even Apple — which carries perfectionism to its limit — ships products that, the moment they land, are not perfect.

What’s more, even if we were willing to push 3ook.com’s launch back a year, that wouldn’t mean we could make a perfect product. Only by keeping an imperfect 3ook.com running can the team watch usage data, gather user feedback, and talk with publishers and independent authors inside a living ecosystem — gradually adding the right features and cutting the useless ones, step by step, feeling out alongside the community what a “third-generation bookstore” should look like.

Don’t take this for high-sounding soft promotion; let me give a concrete example.

3ook.com’s pitch is “giving books a voice.” Audiobooks are expensive to produce — the privilege of bestsellers — but by driving costs down with generative AI, 3ook.com lets any work be read aloud in a voice artist’s, or the author’s own, voice. Without market scale, having humans read niche works line by line can’t recoup the cost; reading in a “niche language” like Cantonese is harder still. And so our centerpiece is Cantonese narration.

But precisely because Cantonese isn’t mainstream — where AI narration in English and Mandarin runs very near 100% accurate — AI reading Cantonese text carries more flaws. Even at 99% accuracy, a 1% misread rate means a 100,000-character text gets read wrong a thousand times. Thick-skinned as I am, I can’t wave that away with a glib “imperfectionism.” More awkward still, the misread character is sometimes crucial — for instance, the 區 in author Au Ka-lun’s surname (read “Au”) getting voiced as the 區 in 區塊鏈, “blockchain” (read “keoi”). How could I face this bestselling author?

Happily, AI narration added a dictionary feature last month: whenever a misread turns up, you can add an entry to correct it. To continue the example, I’ve already added a line to the dictionary:

"區家麟": "(au1)家麟"

explicitly specifying that, in this context, 區 is read “au1,” correcting the AI’s misreading. 3ook.com is fully open-source: not just team members, but any technically inclined user — any Hongkonger devoted to preserving Cantonese — can take the initiative to add an entry to the dictionary hosted on GitHub and file a pull request. Quite literally, this lets the AI learn from its mistakes, sharpening its pronunciation one character at a time.

It isn’t hard to imagine: had we waited for Cantonese narration to be perfect before launching, the result wouldn’t have been a delay of a year or two — it would have been never launching at all, because it would never be good enough. Real users using it and taking part is the precondition for continuously improving and refining the product. Setting aside my ego and shipping flawed Cantonese narration did draw its share of mockery; but every criticism is a user taking part, and every correction, in turn, becomes nourishment for the product.

Inching Forward Against Forces Beyond Control

In the end, imperfectionism seeps into my choices about life, too.

Where I live, justice is wanting and freedom under pressure — even writing an article means looking over your shoulder. Starting over in a democratic, free, at least outwardly more “perfect” system is a reasonable path. I have no wish to judge that choice. And yet: more than a result reached in a single leap, I value the process of taking part. I would rather stay on this imperfect, even ugly, patch of land — and within the shadow of forces beyond my control, with small but forward steps, keep making myself and society a tiny bit better.

A crack — that’s where the light gets in.

p.s. “The New Boy Who Cried Wolf”: Ever since I finished the IBO (Initial Book Offering) of “Attack on Freedom” late last year, I started writing the book. I’ve told people several times that I was “almost done with the first draft” — but the manuscript dragged on and on, and my words turned into lies. Early last week I again told the 100 co-publishers I was “almost done with the first draft,” and then, a few days later, I actually finished it.

p.p.s. No need to spell it out: the first draft of “Attack on Freedom” isn’t perfect. The co-publishers are reading the trial edition right now — letting the light into the text.

Apart from old salts like me who still use it daily, plenty of people wrote WordPress off as outdated long ago. Last week, WordPress 7.0 — codenamed Armstrong, a tribute to jazz legend Louis Armstrong — shipped with a suite of generative-AI features. Is this a has-been uncle dressing up in trendy clothes, all awkward imitation? Or a veteran NBA player who’s just been paired with a hot rookie, suddenly dangerous again?

I’ve gone in first: I upgraded my blog, ckxpress.com, and took it for a spin. Here are my first impressions.

Collaboration Arrives, AI Gets Integrated

WordPress launched the Gutenberg project back in 2017, a four-phase plan to overhaul the core. Phase one was the block editor (the kind that treats text, images, video and other elements as “blocks” — nothing to do with blockchain), rolled out gradually from 2018. Phase two was full-site editing, introduced in early 2022 and steadily strengthened since. Last week’s WordPress 7.0 raises the curtain on phase three: real-time collaboration.

Back in 2005, the startup Writely already supported collaborative, real-time writing; after Google acquired it, the result launched in 2006 as the now-familiar Google Docs. Two decades on, WordPress has finally filled in the collaboration gap — better late than never. The irony, though, is that even though ckxpress.com has run on WordPress for 21 years, it’s always been a one-man band. I have no one to collaborate with. Then again, if I really needed multi-user collaboration, I’d probably have jumped ship long ago.

Of the entire Gutenberg roadmap, the feature I’m most looking forward to — multilingual support — got bumped all the way to phase four, which doesn’t even have a concrete development timeline yet. The saving grace is that whatever the core lacks can almost always be patched in with a plugin. That’s precisely where WordPress’s strength lies. ckxpress.com needs English support, so a little over three years ago I evaluated a few multilingual plugins, settled on Polylang, and documented the whole process.

Polylang only provides the framework — you still have to supply the text yourself. So while choosing a translation plugin, I also evaluated three translation helpers: Google Translate, DeepL and ChatGPT. At the time, ChatGPT had only been out for half a year and the other large language models hadn’t yet found their feet, but it was already enough for me to conclude, beyond doubt, that generative AI was far stronger than the old-school translation software.

Not long after that, whenever I wasn’t being lazy, I’d put out an English version of my weekly newsletter. AI translation may have saved me 99% of the effort, but even so — pasting the article into a chatbot (ChatGPT at first, then Gemini, then Claude), reviewing and revising it (more edits were needed early on, though it kept improving), pasting the text back into WordPress, restoring the images and any formatting lost along the way — it all still adds up to a chore. With a piece like 當《尋秦記》項少龍穿越回秦代搞維穩, I just can’t muster the enthusiasm to do all this work that will probably go unread anyway.

By integrating AI, WordPress 7.0 lays the groundwork to drastically simplify all of the above — handled entirely in the back end.

It’s the Plumbing and Wiring, Not the Appliance

That said, if you think upgrading to WordPress 7.0 means you’ll have an AI assistant standing by, ready to take your orders, you’re in for a disappointment.

True to WordPress’s longstanding philosophy, what it offers is the plumbing and wiring, not the appliance itself. Applied to AI, that principle takes the form of three things:

• Connectors Hub: a central settings area where you enter the API keys for AI services, to be shared across all plugins.

• AI Client: a unified programming interface so that plugins calling AI don’t each have to rewrite their own.

• Abilities API: defines what actions AI is allowed to take within WordPress — reading posts, changing categories, generating summaries, and so on — while controlling permissions.

The Connectors Hub supports Anthropic, Google and OpenAI out of the box; I plugged in my Anthropic API key. For image generation, you can use the latter two.

It’s a sensible, clever design — but a curious one, and as a result it leaves the average user thoroughly unmoved. It’s a bit like when the REST API was first added: ordinary people felt it had nothing to do with them, yet two or three years later nearly every modern plugin, mobile app and headless architecture relied on it. The AI infrastructure will probably follow a similar path: no one feels it when it first arrives, but after a while the AI plugins across the ecosystem will all become more usable, cheaper and more integrated thanks to this foundation.

Still, before the plugins you actually use add support for the AI infrastructure, you can dip a toe in via the official AI plugin. Once you’ve filled in the API key and installed it, you can switch on whatever you like in the settings — title generation, cover images, summaries, and more. For testing purposes I turned everything on; here are a few of those features, tried out directly on Chinese version of this very article.

• Alt-text suggestions: regular readers don’t see alt text, but it helps search engines. Filling it in image by image is a pain, so I usually leave it blank — exactly the kind of thing best left to AI.

• Title suggestions: useful but basic. If I genuinely wanted AI’s input, I’d ask it for several at once, then pick and tweak.

• Tag suggestions: not bad. I might keep using this one.

• Paragraph expansion: beyond padding out a word count for a fee, I have no idea what this is for. Who actually writes like this? I’d have to seriously consider cutting them off.

There’s a whole pile of other features, and honestly, most are useless to me. I resist leaning on AI during the actual writing process — and yet the one feature I need most, translation, is the one it doesn’t offer. But that’s no fault of its own: the AI plugin ought to stick to basic editing functions. Once plugins like Polylang add AI translation, maintaining a multilingual site will become a breeze.

As infrastructure, what WordPress 7.0 provides is an AI foundation. It may not help much for now, but as plugin developers ship updates and the Abilities API gets put to wider use, WordPress will be able to keep up with the times and endure.

p.s. While WordPress is busy adding AI infrastructure, over on the other side Cloudflare simply vibe-coded an entirely new content management system called EmDash. Following company tradition, it was released on 1 April — but it’s no April Fools’ joke; it’s positioned as the “spiritual successor to WordPress.” I’ve always admired Cloudflare’s products, but I’m not about to run experimental new software on my own blog.

Tomorrow is the 16th anniversary of Bitcoin Pizza Day. I reread the three pieces I’ve written on the topic over the years — nothing to add, nothing to correct. Rather than rehashing the same tune, I’ll just relink the old pieces for anyone interested:

• We Are All the Protagonists of Bitcoin Pizza Day

• Thinking Fast and Slow on Bitcoin Pizza Day

• Bitcoin Pizza Day, 15 Years On: The World Has Learned Not to Buy Pizza with Bitcoin

Besides, with Bitcoin’s price in the doldrums right now, writing about Bitcoin Pizza Day probably isn’t what most people are in the mood to read. The topics I want to write about and the topics readers are interested in are often separated by a chasm.

Last week’s newsletter, on the absurdity of Hong Kong’s ticketing systems, was a piece of grumbling I hesitated a long time to write — and it racked up the most Comments, Likes, and Shares in the newsletter’s history. Truly “CLS”. Being detached from the ground is meant to be a feature of me, not a bug. But I get it — caring about local, on-the-ground topics is simply human nature.

In last week’s P.S. I mentioned my eMPF experience, and quite a few readers wrote in asking for more. So here, by popular demand, is the even more granular grumble.

First, a quick primer on eMPF, especially for overseas readers who haven’t yet been driven away by Hong Kong-specific content.

MPF stands for Mandatory Provident Fund. Launched in December 2000, it’s exactly what the name says — mandatory. Employers and employees in Hong Kong are each required to contribute 5% of monthly wages, which can be withdrawn after age 65.

eMPF, then, is a wholly-owned subsidiary of the Mandatory Provident Fund Schemes Authority, which became operational in June 2024. While MPF forces employers and employees to contribute, eMPF forces the banks and insurance companies that manage MPF schemes to use its system. Per the official website, “The eMPF Platformcentralizes and digitalizes all MPF schemes’ administrative processes, allowing users to manage MPF online at ease.”

Sounds beautiful. The reality is this: before HSBC — my MPF provider — was forced onto eMPF, I had been using my regular HSBC online banking account to handle everything in the same interface with just a few clicks. After being forced to switch to eMPF for no good reason, I became one of countless Hong Kong victims — two trips to the bank, multiple calls to customer service, several new accounts opened, and a wait of over two months. After all that back-and-forth, I finally completed my first successful contribution.

Here’s my experience log. Any resemblance to your own story is purely unfortunate.

• Late February: Planning to make a tax-deductible voluntary contribution before the financial year ends on March 31, I found the original HSBC MPF interface was no longer accessible.

• Early March: To register for eMPF, I was first forced to sign up for “iAM Smart” — something I had been deliberately avoiding. During the eMPF registration, the password generated by Proton Pass was rejected for being “too long”. One small saving grace: at least they didn’t force me to install a mobile app — but I had to install “iAM Smart” first, which is effectively the same thing.

• March 5: I initiated a contribution on the eMPF website and received an invoice. Different MPF providers support different payment methods, and after rereading the instructions where “I understand every word but not the sentence,” I still couldn’t tell what payment methods I could actually use besides a cheque.

• March 5: Logged into HSBC online banking, selected “Pay Bills,” and could not find the MPFA or HSBC MPF in the payee list.

• March 6: Asked AI. Gemini told me to call customer service. Figuring that intra-system problems are best answered by intra-system AI, Deepseek told me I could just transfer the money to AIA for collection. Thank goodness I didn’t follow that advice.

• March 6: Called several MPFA-related customer service lines. After being bounced around like a human pinball, I was told the safest method was a cheque, or that I could ask HSBC if there were other supported payment methods.

• March 6: I do have an HSBC cheque account, but I hadn’t used it in over a decade(?) and couldn’t find the chequebook. I asked AI which banks could issue e-cheques and learned that only Bank of China, CMB Wing Lung, and Fubon Bank support them — HSBC personal customers can only deposit e-cheques, not issue them.

• March 7: On the rare Saturday off, I went to the MPF counter at an HSBC branch to ask. They confirmed there was no other option — cheque only — and even suggested I mail it (oh god). The bank staff, by the way, seemed pretty resentful of eMPF, exuding a “we’ve been sidelined, what can we do” vibe.

• March 9: Activated the e-cheque function on my dormant Bank of China account, transferred money in, issued an e-cheque, and uploaded it to the eMPF website. The upload page had a bug, but I worked around it and submitted successfully. Anyone who knows me a little can probably imagine how out-of-character the scene of me issuing a cheque was — but to survive in Hong Kong, I compromised, one indignity at a time.

• March 10–19: Assumed it was done. Forgot about it.

• One day in March: Received a physical chequebook from Bank of China. A waste of planet resources — apparently activating e-cheques automatically gets you physical ones too. Such warmth.

• March 20: Received an email from eMPF reminding me to make a contribution. Taking 11 days to process an e-cheque is a bit much, but whatever, I’d wait and see — I no longer had the energy to call customer service.

• March 27: Received a final notice from eMPF reminding me about the contribution. -.-! Done playing. I figured, forget it — I’m not obligated to make voluntary contributions anyway, and given how much money Just Books has lost, maybe I’m not even qualified to be paying taxes.

• March 27: Logged into my Bank of China account to withdraw the funds I had deposited — only to discover the cheque had been cleared on March 23. WTF, is this a scam?!

• April 9: Received an email from eMPF saying the contribution was complete, but logging into the website still showed no transaction, and I had no way to tell whether it had cleared before or after March 31 — i.e., which tax year it counted toward. Mentally exhausted, I let it go.

• May 15: Received an email from eMPF, finally confirming that the tax-deductible voluntary contribution was processed in the previous financial year. From initiating the contribution — thinking it would take a few clicks — to receiving confirmation, it took a full 71 days.

Readers may have an appetite for local content, but writing up this account leaves me agitated and conflicted. There are so many more important issues — why am I venting about these trivialities?

More importantly: MPF is a basic safeguard for people’s retirement. The infrastructure was already built, working effectively for years. Why does the government insist on starting from scratch — centralizing what was once distributed across individual banks, and replacing it with a system that feels 20 years behind?

P.S. Last week I talked about the absurdity of buying tickets to Sandy Lam’s concert. This week, the joy of actually being there.

From the moment it started, Sandy — looking like some kind of mind-body-spirit guru — lay down and gently sang “Breathe In… Breathe Out.” Then, without a single word, she sang 18 songs nearly in one breath. Only before the “final” song, “Goodbye Sorrow,” did she finally speak. Choking up, she said she had been holding back tears for a long time. Throughout the tour, people had constantly asked when she would perform in Hong Kong. The delay had been because Kai Tak Stadium was hard to book — and Hong Kong was the stop she most cared about, the one she had been looking forward to the most.

“Not really a stop — because this is my home.” That line of Sandy’s hit me right in the heart.

The concert was officially titled Resonance: Reconstructed Hong Kong Exclusive. I hadn’t paid much attention to the title, but after the show I realized — unlike the cookie-cutter touring concerts, this one was designed specifically for Hong Kong. One segment recreated Sandy’s childhood home in North Point, with a tram passing by. The setlist, too, was tailored for Hong Kong fans: after opening with an English song, the next 19 songs, plus two encores totaling 7 songs, were all Cantonese — all freshly arranged by Anthony Lun.

It’s not that I’d feel cold without Mandarin songs — I’m not that fragile. Honestly, I would have loved to hear live renditions of “At Least I Have You,” “When Love Becomes the Past,” and “I Heard Love Has Returned.” But Sandy, with 40 years of work behind her, has such a vast catalogue that to fit in early-career songs that were never popular in Taiwan or the mainland — “Love Doesn’t Age,” “Day and Night,” “You Are My Man” — the kind whose mere intro gives me goosebumps — meant giving up some of the songs from her overseas-development years.

This Cantonese setlist, belonging to Hong Kongers, wasn’t about expressing “what we are not.” It was about defining who we are through the work itself, singing into the hearts of the audience and producing Resonance.

I’ve never been so moved by a concert. Thank you, Sandy.

Whether anyone should be held responsible is for the independent commission to determine; the public should not rush to judgment. While we wait for the findings, however, it may be worth turning to another case — one that shows just how proactive the Food and Environmental Hygiene Department (FEHD) can be when it sets its mind to enforcement, and which may serve as an instructive example for other government departments.

Hong Kong law requires anyone running a “place of public entertainment” to hold a licence, with criminal penalties for failure to do so (Cap. 172, s. 4). On 29 June last year, Book Punch — an independent bookstore in Sham Shui Po — hosted a graduation showcase for the third cohort of its in-house stand-up comedy class. The event ran across two sessions, pay-what-you-wish. No licence had been obtained. Yesterday, the bookstore and its owner were each found guilty on two counts and fined HK$3,000 per summons (case nos. KCS37515 & KCS37516/2025).

I attended both the trial in the morning and the verdict in the afternoon at Kowloon City Magistrates’ Courts. Watching how this case was successfully built and prosecuted, I noticed at least three commendable details that other government departments would do well to study.

Proactive Enforcement: Scrolling Facebook in the Midst of a Busy Schedule

According to the Agreed Facts, two FEHD Health Inspectors I came across a social media post by Book Punch about the upcoming event on 25 June 2025, and from that moment began their investigation.

In other words — assuming I have understood correctly — between their routine inspections and their handling of public complaints, FEHD officers nevertheless found time in their busy schedules to spot, on Facebook, that an independent bookstore was hosting what looked like a public entertainment event without the relevant licence, and to swing into action. This kind of proactive, public-safety-minded enforcement stands in stark contrast to the popular stereotype of bureaucracies — that they go strictly by the book, that doing less means erring less. It is a model worth emulating across the rest of government.

Infiltration and Cunning: Undercover Agents, Cash in the Cardboard Box, Floor Plans from Memory

The Agreed Facts go on to state that the two FEHD inspectors signed up via the registration link in the Facebook post. On the evening of the event — 29 June last year, around 6:00 p.m. — the two inspectors arrived in plain clothes. There were about 50 audience members. Six performers took turns delivering jokes and using physical comedy, to frequent laughter from the audience. At 8:00 p.m., after a group photo, the event came to a close.

According to the case, the two undercover inspectors not only documented the event in detail and took 11 photographs — they each also dropped HK$100 into the cardboard box being passed around, like the moles in classic Hong Kong cinema who, to win the trust of the boss, are willing to dirty their own hands by committing the very crime they are investigating. The day after the event, the two inspectors drew, from memory, the floor plan of Book Punch — the key piece of evidence that cracked the case.

Those of us raised on Hong Kong cinema, with its long tradition of cop-and-mole thrillers — the kind that gave us Infernal Affairs, later remade by Scorsese as The Departed — tended to assume undercover work was reserved for triads and drug lords. It turns out it is also reserved for stand-up comedy nights.

Statutory Interpretation: A Bookstore Without a Stage Is a Stage After All

The prosecution and defence agreed on what happened. The dispute was over what the law meant.

Mr. Lawrence Lau Wai-chung SC, for the defence, argued that the Ordinance regulates “stage performance” — not “performance” as such. Citing both the Oxford English Dictionary (“a raised floor or platform… on which actors, musicians, etc. perform”) and the Cihai Chinese dictionary, he submitted that since Book Punch had no stage, what took place that day was not a “stage performance” within the meaning of the Ordinance. The narrowest interpretation, he argued, should apply — both because every word in a statute is presumed to have meaning, and because cultural activities enjoy constitutional protection under Hong Kong’s Basic Law.

After more than four hours of deliberation, Magistrate Mok Tsz-chung was unpersuaded. The legislation, he ruled, exists to protect public safety; reading it too narrowly would defeat that purpose. So long as a performance draws an audience, whether or not there is a literal stage is beside the point. He convicted on both summonses.

The Wang Fuk Court fire killed 168 people. Even though residents had complained beforehand about contractors colluding on tenders, district councillors collecting large numbers of authorised votes, scaffolding netting failing to meet standards, flammable enclosures, and construction workers smoking on site, no one was prosecuted on any of those counts and the disaster was not prevented. Some have accused various government departments of passing the buck and going through the motions. Yet the Book Punch case demonstrates so amply the initiative and execution of these very departments that any criticism of government incompetence collapses on its own.

Don’t take a single case as too small to reveal what these departments are capable of. This is, in fact, the third successful prosecution of Book Punch — after earlier convictions for selling alcohol without a licence and operating an unlicensed school (the latter for hosting a Spanish-language interest class). Each time, in matters that ordinary citizens would consider harmless trifles, the FEHD, the Education Bureau and others have managed to unravel the threads, descend on the premises by surprise, and — twice now — send in undercover agents. The judiciary, for its part, has played its part in securing convictions every time.

A fourth charge is pending. In March, Book Punch was raided again — this time by the National Security Police — over its sale of The Troublemaker, a biography of Jimmy Lai, the jailed pro-democracy media mogul whose own trial under the National Security Law has drawn international attention. Pong Yat-ming and three staff members were arrested for “knowingly selling a publication with seditious intention” and released on bail. Given the impressive performance government departments have shown in recent months, and their evident determination to safeguard public safety, what kind of future remains for Book Punch — and for Hong Kong’s other independent bookstores — is something the public can simply look on and see.

Back when Facebook was still the place to be, posts like this were everywhere — people speaking, half in jest, to the algorithm itself.

At first everyone understood it was just a figure of speech. The post was really aimed at friends who might see it and chime in with recommendations. Social networks, after all, were where people talked to people.

But a decade and change later, users seem to have forgotten the rhetorical sleight of hand. The phrasing has gone from joke to habit — we now treat the algorithm as if it were a person. Late last year, the joke even came full circle: Meta picked it up and rolled out an experimental Threads feature called Dear Algo, openly inviting users to address the algorithm directly.

So what is an algorithm, really? And how does it relate to AI? Here’s how I think about it.

Bob’s Recipe

Long ago, I studied Computer Engineering at a Chinese University of Hong Kong. It was a traditional degree at a traditional school, with the usual core requirements: C Language, Data Structures, Operating Systems — and what we now hear thrown around constantly: Algorithm.

Loosely speaking, an algorithm is a method or sequence of steps for solving a particular problem. In that broad sense, even a recipe is an algorithm — its problem being how to produce a given dish. But for a Computer Engineering student, the definition is tighter: a finite, unambiguous, executable sequence of instructions for solving a well-defined problem.

Imagine you have a printed Yellow Pages (Facebook’s prehistoric ancestor) and you need to find the number for “John Wick.” The most intuitive algorithm is linear search: start at “Aaron Au” and go down the list, name by name, until you hit your target. It works, but it’s a bit dim.

A smarter approach is to start in the middle. If you land on “Nancy Chan” — which comes after “John Wick” alphabetically — you halve the remaining range and try again at the 25% mark. Repeat until found. This is binary search.

In the classical sense, every part of an algorithm — its steps, its inputs, its outputs — is crisply defined. The user states the problem clearly; the programmer designs the steps; the user supplies the input; the algorithm produces a specific output. Every detail is traceable.

Carol’s Eyes

Now picture this. Alice, a manager, asks Bob — her IT guy — to find all the photos containing cats among ten thousand images.

To Alice, the requirement seems perfectly clear. To Bob, it’s anything but. How do you tell a computer what a cat is? How do you define one? It’s a hopelessly fuzzy specification — but if Bob went back to ask Alice for clarification, he’d probably get fired. Funny enough, if Alice’s young daughter Carol asked her the same question — what is a cat? — Alice would discover that while she can recognize one instantly, she can’t actually explain how.

She’d find that listing features — four legs, fur, whiskers — never quite works, no matter how exhaustively. Either the definition is too loose and includes dogs, or it’s too strict and fails to recognize a cat curled into a ball. And yet, the moment Alice shows Carol a few photos and points out which ones have cats, Carol picks it up almost instantly.

This is how every one of us learned. It’s the kind of learning humans have long taken pride in — the very thing we believed put us above machines. Until AI matured.

Back when I was a student, AI didn’t yet mean “any technology that makes computers seem smart.” It had a much narrower definition, and it had to involve machine learning. Machine learning gave computers something resembling a human capacity: the ability to handle problems we ourselves can’t fully articulate, problems no traditional algorithm could solve. “Find the photos with cats” is now trivial — not because we got better at defining the problem, but because computers, fed enough examples, learned to recognize cats the way a child does.

Alice’s Mystery Box

Now imagine Alice has moved on to Meta, where she leads the Threads team. She turns to Bob with a new request: from the hundreds of millions of posts flowing through the platform every day, generate a feed that’s maximally addictive.

As before, Alice can name the goal but can’t describe its features. So Bob falls back on signals — comments, likes, shares — to filter for whatever seems to be hooking users. This was the early algorithmic feed: still classical in spirit, with explicit rules and conditions, predictable outputs for given inputs. Tweak the rules, and both engineer and user could steer the results.

But as AI grew more powerful, Alice no longer needed to articulate what makes content addictive. Bob could let the AI figure it out and serve up the most habit-forming feed possible, herding users like livestock. The word algorithm is still in use, now meaning “the logic by which a social platform filters information” — but its meaning has shifted entirely. The classical algorithm was a recipe Bob wrote out for the computer; every step was visible. The AI-era algorithm is a black-box chef Bob has trained, and even Bob can’t say why it seasons things the way it does.

For us users, the situation is worse still. Beyond knowing that ads will be slipped in somewhere, we have no real idea how the feed is being cooked. Whatever theories we have are pure guesswork. The irony is that more and more users are happy to remain in the dark — they don’t bother checking the nutrition label, and have started to enjoy the mystery of the surprise box.

We tell ourselves we’re sitting at a fine Japanese omakase counter, paying a premium so a master chef can hand-pick ingredients and prepare each course with care. We forget we’re actually scarfing down free fast food. Why would the chain bother sourcing real ingredients when it isn’t charging us a cent? Nutritionally, though, you can rest easy: the feed is guaranteed to be rich in dopamine. You’ll find it delicious.

Dear User, this is the algorithm.

P.S. Another casualty of the algorithm is the way it crams certain eateries to bursting while leaving worthy little shops deserted. Sometimes the packed-out spots are nothing special, just blessed by the Xiaohongshu gods. Other times the favored places really were excellent — until the crowds arrived and both food and service collapsed, never to recover. Living in Hong Kong, steering clear of the algorithm and seeking out independent shops on your own is a basic skill for eating well and living well.

Bitcoin is the perfect example. At the end of 2024 it broke through $100,000 for the first time, and everyone was either scrambling to buy or thinking it had risen too high and vowing to buy on the pullback. As I write this, Bitcoin is trading around $75,000 — up 20% from the $62,000 low, but still more than 20% below $100,000, and a full 40% below the all-time high of $126,000. If you believe in Bitcoin, putting spare capital to work here should be a simple decision. Yet instead, what we focus on at this moment is the risk — like the arrival of quantum computing. This was especially true after the Google Quantum AI team, together with Stanford and Ethereum Foundation researchers, published a white paper last month showing that the quantum resources needed to break Bitcoin wallet encryption are nearly 20 times less than previously estimated, suddenly putting the quantum threat squarely in the spotlight.

I agree the quantum computing risk is real, but I’m sticking with my advice in Buy $100 of Bitcoin Every Day — with one caveat. Let me first set the context, then offer concrete advice on how Bitcoin holders should prepare for the arrival of quantum computing.

Quantum Computing: A Y2K Bug Without a Set Date

Last century we had the urban legend of Y2K: the moment the clock struck midnight on January 1, 2000, banking systems would crash, public services would grind to a halt, airports would descend into chaos, and planes might even fall from the sky. In the end, Y2K was all thunder and no rain — apart from some elderly folks being tricked into buying deworming pills, nothing happened.

Quantum computing, by contrast, is a genuine Y2K bug. It won’t just affect Bitcoin, Ethereum, and every other cryptocurrency — the entire internet, all banking, and military communications will be impacted, and everyone needs to prepare in advance.

What makes this tricky: twenty-some years ago, everyone knew exactly which second Y2K would arrive (assuming it didn’t just stand us up). This time, the “quantum Y2K” has only given humanity a vague appointment called Q-Day — Quantum Day, the moment a quantum computer first has enough power to break mainstream public-key cryptography like RSA and ECDSA. But estimates for when Q-Day will actually arrive range anywhere from a few years to a few decades.

Bitcoin is secure because only the private key can move the assets, and deriving a private key from a public key through computation would take longer than the history of humanity — longer even than the universe. That logic holds for today’s computers, and it already accounts for computing speeds increasing by factors of thousands, billions, even trillions of times. But quantum computers work on an entirely different foundation, making them exceptionally good at deriving private keys — rendering mainstream public-key cryptography fundamentally broken. In other words, any Bitcoin stored at an address whose public key has ever been exposed — estimated at 6.7 million BTC, or 34% of circulating supply — could be easily stolen after Q-Day.

BIP-360: Voluntary Migration to a Quantum-Resistant Address

To counter this risk, the Bitcoin community has drafted BIP-360: Pay-to-Merkle-Root (P2MR). If adopted, holders will be able to move their Bitcoin into a new wallet format, P2MR, that doesn’t expose public keys (sparing you the technical details) — so even when Q-Day arrives, there’s nothing to worry about.

BIP-360 effectively counters quantum attacks, but it requires holders to proactively move their Bitcoin to the new safe wallet. Here’s the problem: of the 6.7 million BTC sitting in quantum-vulnerable addresses, an estimated 3.7 to 4 million BTC have lost private keys — including the 1 million BTC held by Satoshi. In other words, without some other solution, there is nothing to do but wait for Q-Day to arrive, at which point these coins will be “mined” by whichever institution or nation-state gets a working quantum computer first.

Given corporate competition and geopolitical rivalry, it’s unlikely that multiple companies or multiple countries will simultaneously possess the first practically deployable quantum computer. Which means that after Q-Day, a large amount of Bitcoin will end up in the hands of a few — or even a single — organization.

Of course, the real game theory is more complex. If Q-Day arrives suddenly, Bitcoin will plunge, so shorting heavily first and then proving to the market that quantum computers have arrived could be even more profitable, and faster. Or consider this: relative to Bitcoin’s “mere” $1–2 trillion market cap, using a quantum computer to steal US dollars or military secrets would be far more valuable. Either way, as quantum computers arrive and proliferate, quantum-vulnerable addresses will become ATMs — and more fundamentally, Bitcoin’s entire narrative of decentralization and digital gold could collapse.

BIP-361: Forced Freezing of Unmigrated Assets

And so, recently, the Bitcoin community has proposed BIP-361: Post Quantum Migration and Legacy Signature Sunset. It takes the active approach: three years after activation, the consensus layer will force legacy quantum-vulnerable addresses to only send, never receive Bitcoin; then five years after activation, any Bitcoin that still hasn’t migrated will be frozen, preventing post-Q-Day plunder.

You don’t need me to explain that while BIP-361 is a proactive bomb-defusing measure, forcibly freezing holders’ assets is undeniably an abandonment of principles the Bitcoin community has upheld for over a decade. Compared to the simple directness of BIP-360, BIP-361 is enormously controversial and stands little chance of winning clear community support. Freezing specific addresses violates core values; not freezing them means being picked clean. And with no concrete date for Q-Day, no hard deadline, reaching community consensus is all the more difficult.

BIP-360’s activation via soft fork is only a matter of time, not really in doubt. BIP-361 or its variants, however, will almost certainly struggle to get Bitcoin Core developers on board for a long time. Yet at the same time, there will no doubt be plenty of supporters — especially corporations holding large amounts of Bitcoin, who aren’t as “stubborn” as the fundamentalists and tend to take a more “pragmatic” approach. With both sides deadlocked, a hard fork becomes very likely — splitting into a die-hard BTC and a pragmatist coin like BTQ. Or the other way around: the pragmatist chain may inherit the OG name BTC, while the die-hards become BTCC, Bitcoin Classic — after all, ticker symbols are ultimately decided by the exchanges.

A split is usually seen as a negative, but in the crypto world, a split is sometimes an “airdrop” opportunity. Take Ethereum’s hard fork after The DAO hack. Because ETH — the side that forcibly froze assets — had the backing of the Ethereum Foundation, Vitalik, and most core developers, its price was unaffected by the fork, rising from $12 in July 2016 to around $2,400 today. Meanwhile ETC — the die-hard Ethereum Classic — went from $0.75 at launch to $8.70 today. ETC may look shabby on the surface, but its market cap still sits at $1.3 billion — all of it an “airdrop” to former ETH holders.

The One Thing Bitcoin Holders Need to Do

After all that, what do Bitcoin holders actually need to do in the face of the quantum threat? Self-custody.

The reason is simple. When Bitcoin really does split in two, existing holders will own both coins. If you hold 1 BTC before the hard fork, afterward you’ll have 1 BTC + 1 BTCC. But you must control your own private keys to receive the forked coin. If your Bitcoin is on an exchange — or even in an ETF — the custodian will likely pick a side. Your original holdings and units won’t be affected, but you’ll probably never receive the new coin.

If the two post-split coins turn out to be as lopsided as ETH and ETC, and your exchange picks the right side, your loss is relatively limited — you’ve simply handed your “airdrop” to the exchange for free. And in fact, this has already happened repeatedly during Bitcoin’s 2017 and 2018 hard forks into BCH, BTG, and BSV.

The Ethereum precedent offers a reference, but if Bitcoin really does hard fork, the price ratio won’t necessarily be as one-sided as ETH and ETC. This time, core developers will likely side with the die-hards, and the assets being frozen aren’t a hacker’s — they belong to OGs including the founder himself. One side controls the capital, the exchanges, the ETFs; the other side holds the moral legitimacy. It’s evenly matched. In that scenario, being stuck with only the coin your exchange chose becomes a significant cost. The “fun” part: the forked-out coins don’t disappear — they all become the exchange’s assets.

The oft-repeated saying “Not your keys, not your coins” — this is exactly what it means.

p.s. Has anyone noticed that since the government loudly declared last year it would “make Hong Kong a virtual asset management hub,” Web3 has hardly been mentioned lately? Of course — prices are down, AI is hot, the Northern Metropolis is here. Time for another new story.

I have rarely attended concerts in recent years. Without checking social media, I often miss the news, and even if I knew, tickets are too hard to buy. Entering middle age, I have lost the drive to “hunt for tickets” just to get in. However, after making an exception last month for Stefanie Sun’s Just Before Sunset tour, it felt like opening Pandora’s box, and I began paying attention to other concerts again. One morning, while listening to Vani’s Choice on Commercial Radio, I learned that Shino Lin would be coming to Hong Kong for a concert on April 4, and I bought a ticket without hesitation.

In 2021, a young person in mainland China posted on Weibo: “Recommending a niche Singaporean Mandarin female singer—Stefanie Sun,” which became a hot topic among “elders.” Some mocked the ignorance of the poster, while others lamented the changing trends. Five years later, Stefanie Sun’s world tour came to Hong Kong, and tickets were incredibly hard to find. After much effort, I managed to buy a ticket and squeezed into the Kai Tak Main Stadium for the first time. I discovered that Stefanie Sun being “niche” was fake, the “home” court was fake, and even asking the “Hong Kongers” in the audience for requests was fake; the only thing real was that the species known as “Old Hong Kongers” becoming niche in their own home court.

There was nothing wrong with the quality of the concert. Just Before Sunset was the real deal—Stefanie Sun’s singing, the production of the show, and the facilities of the venue were all first-class. The whole experience was like having a buffet in a five-star hotel (crowded with mainlanders): very expensive, very high-end, and the food was fresh and delicious. The only problem was that I am still more accustomed to and prefer Cha Chaan Tengs and Dai Pai Dongs.

This Shino Lin concert was titled Tempting Heart Coordinates, and the poster displayed “East Longitude 114.1372, North Latitude 22.3453.” The promotion said those were the coordinates of the venue, Queen Elizabeth Stadium. It sounded reasonable, but being curious, I checked repeatedly and found that those coordinates actually point to the Mei Foo area. The coordinates for the stadium should be “East Longitude 114.1788, North Latitude 22.2753.” I don’t understand why the organizers made such a low-level mistake, but I suspect there wasn’t much of an “organizer” at all. Not only was there little promotion, but the poster also looked like it lacked a production budget. Holding it during a long holiday when half of Hong Kong had left was an even more ill-advised arrangement. It’s a great pity.

Compared to Stefanie Sun’s concert, which held two shows at the 50,000-seat Kai Tak Main Stadium with tickets selling out instantly, Shino Lin’s concert was a single show at the 3,500-seat Queen Elizabeth Stadium. Tickets were not only much cheaper but could be bought at any time, and by my observation, a third of them remained unsold, leaving large patches of empty seats. However, the resonance and emotion brought by Tempting Heart Coordinates stood in stark contrast to its scale and production budget.

In this era of fragmented audiences dominated by short videos and algorithms, where even Stefanie Sun can become a “niche singer,” I think it’s appropriate to “popularize” who Shino Lin is before I go on.

Born in 1973, Shino Lin began singing in Japanese hot spring hotels at the age of 17 under the stage name Shino. After returning to Taiwan in 1994 to sing in pubs, she debuted in 1998. Her first album, Shino Lin, was an instant hit, selling 700,000 copies. With a style rooted in rock, Shino made no effort to hide the large tattoo on her upper right arm and later admitted to working in a hotel to make ends meet—she was considered very forward-thinking at the time. Over the next few years, albums like She Knows and This is Shino 3 continued to sell well, and she held large-scale concerts. Just as her career was at its peak, she was diagnosed with bipolar disorder in 2003 and temporarily left the music scene.

An even bigger turning point was yet to come. Originally scheduled to release a new album in late 2007, Shino committed a grave mistake mid-year by driving without a license while intoxicated and hitting a nurse, resulting in a fatality. She was sentenced to six months in prison (suspended for two years) and reached a settlement of 8 million TWD with the family. Although Shino continued to have occasional singing and performing opportunities, she remained largely quiet overall. She could never return to her former popularity. As she revealed in recent interviews, in addition to spending 11 years paying off the 8 million in installments, she has blamed herself for over a decade, not allowing herself as the perpetrator to be happy.

With this background supplemented, it explains why I am so forgiving of the mistakes on the poster and the details of the production—because rather than “consuming” Shino, I am here to support her. Rather than demanding a visual and auditory experience commensurate with my payment, I hoped more to see Shino pick herself up, walk out of the shadows, and perform with the same confidence she had over twenty years ago. To this, Shino succeeded. I believe the over two thousand fans present could feel it and were moved by her.

Furthermore, a concert is not necessarily better just because it is larger. A smaller venue, shorter distance, a simple stage, no dancers, and no environmentally unfriendly glow sticks were not deficiencies to me; they were pluses. After all, I just wanted to appreciate a singer performing live with full concentration.

Regarding the song selection, Tempting Heart Coordinates was solid and steady. Although Shino has been in the industry for a long time, her peak was followed by nearly twenty years of silence. Her popular repertoire isn’t vast—just enough to support a full concert without leaving fans lamenting that many signature hits were missed. Familiar songs like “Her Tears,” “So What,” “She Knows,” “Irritated,” “Doll Loves the World,” “This is,” “Music All Night,” and “Slow”—a Mandarin cover of Beyond’s “Cold Rainy Night”—were all included as expected. It ended, unsurprisingly, with the most popular song “Tempting Heart,” satisfying the fans’ expectations.

What surprised me slightly was my own memory. Although I haven’t listened to much music in recent years, as soon as the singing started, I realized I hadn’t forgotten anything. I remembered all of Shino’s songs and lyrics perfectly.

The part of the concert that best showed Shino’s heart for her fans and her respect for the Hong Kong music scene was the Cantonese song segment in the second half. After thanking the two guests, Sammi Cheng and Joey Tang, Shino stated she wanted to give back to Hong Kong fans with songs from her idols. she began performing Anita Mui’s “Bad Girl,” followed by consecutive performances of Leslie Cheung’s “Sleepless Night,” Sandy Lam’s “Tilt,” Sammi Cheng’s “Every One is a Super Star,” and Alan Tam’s “Love Trap”—a total of five Cantonese songs from the 80s and 90s. She invited the fans to sing along, pushing the atmosphere to a climax.

Many foreign singers come to Hong Kong for concerts. To please the fans, they often say a few words in “broken” Cantonese or express how much they love Hong Kong food. Hong Kongers, understanding themselves as a minority, are quickly satisfied. Occasionally, some singers like IU sing a Cantonese song every time they visit—from “Wedding Invitation Street” to “Solitude”—which is already over-delivering. But someone like Shino, who already has an original Cantonese song “Doll Loves the World” and sang “Looking at the World through a Doll’s Eyes” with Sammi, and then practiced an entire segment of Cantonese songs to pay tribute—in my knowledge, this is unique.

To be honest, if Shino’s career had been smooth sailing and she had broken into the mainland market, I don’t know if she would still, for the sake of a mere two thousand Hong Kong fans, practice multiple Cantonese songs. Just as we don’t know if Anita Mui or Leslie Cheung were still alive, whether they would be like some artists today, forgetting their original intentions for the sake of the market and abandoning the people of Hong Kong.

I will not dwell on these hypothetical questions that have no answers. The point is, right here today, these singers, actors, directors, creators, and artists before us do not just single-mindedly pursue the largest market. They choose to walk with Hong Kongers, using our language to tell our stories. That is enough.

p.s. Besides the concert, I also went to see Gamer Girls and We’re Nothing during the holiday. I liked both very much; the latter is my favorite Hong Kong film in a year. Herman Yau, who self-financed the production, is very sincere and has incredible execution. After Gamer Girls ended, one of the directors, Bassetto, rushed into the theater for a stage greeting. It wasn’t a traditional Q&A session; she just said it was already the fourth week of screening with very few sessions left and thanked everyone for taking the time to support. Her speech, which lasted less than a minute, was neither impassioned nor intentionally sentimental, yet you could hear a slight tremble in her voice. It seemed as if the audience filling half the small cinema was enough to move her to tears.

I don’t personally use Instagram DMs, but given its massive user base, I expected at least some outcry. Instead, beyond the usual voices from “privacy hardliners” like Proton and Mozilla, there hasn’t even been a ripple on social media. I couldn’t find a single complaint, let alone any talk of a boycott.

I’m not just talking about a lack of discussion within my own echo chamber or the algorithm failing to show it to me. Even when I deliberately searched for it, I found nothing in the Chinese-speaking world. Not only was there no search result, but I actually found many older videos teaching people how to turn off end-to-end encryption.

While my view on privacy remains unchanged, I must admit I’ve had to reflect: Meta is simply satisfying human nature and providing exactly what users want. However, after that reflection, the work of public outreach must continue to ensure that user choices are made out of free will, rather than ignorance or information asymmetry.

Encrypted for everyone… except me

First, let’s re-explain the concept. End-to-end encryption, as the name suggests, means that a message is encrypted throughout its journey from the sender to the receiver. In other words, only the sender and the receiver can read it. The intermediaries responsible for passing the message—including Meta itself—cannot read the content.

The concept of E2EE is easy to understand, but many people are confused about which communication tools actually use it and which do not. The classic example is Telegram. It has a very positive image, as if its privacy is very high, but in fact, general chats—the vast majority—do not have E2EE.

In most cases, when a service like Telegram is not end-to-end encrypted, it’s not that the service isn’t encrypted at all, but rather that the encryption method isn’t end-to-end. It’s “encrypted for everyone, except for me [the service provider].” This is the case for Telegram, for WeChat, and soon, it will be the case for Instagram as well. Many people choose to believe that Telegram will defend user privacy. But what about Tencent? What about Meta?

People who believe Telegram is very private are trusting that Telegram won’t—not that it can’t—decrypt their messages. I don’t question the integrity of Telegram’s founder or its team. The problem is that we cannot rely on the “backbone” of specific individuals to protect user privacy. Any company or organization registered in any country has a legal obligation to cooperate when law enforcement or courts demand it. Even if a founder stands firm, doing so puts them in personal danger. Furthermore, when a company holds the keys to decrypt user messages, those keys could leak due to procedural errors, hacking, or various other reasons.

Regardless of how reliable a founder is, server-side encryption is inherently a bad design. To truly protect user privacy, communication tools must provide end-to-end encryption, and it must be the default.

From the F8 Vision to Removing Encryption

Here’s a joke. In 2019, at the Facebook developer conference F8, Mark Zuckerberg gave a grand speech declaring “The future is private.” He later published a long article titled “A Privacy-Focused Vision for Social Networking,” stating, “I believe the future of communication will increasingly shift to private, encrypted services” and “End-to-end encryption prevents anyone—including us—from seeing what people share.” And so on.

If you took that seriously, you’ve already lost.

Meta’s official reason for turning off E2EE for Instagram DMs is that the feature’s usage rate is extremely low. To be honest, I believe it’s a fact that few people use it, but this is a case of confusing cause and effect. Precisely because Instagram (and Telegram) do not have E2EE as the default, and users have to manually enable the option for specific chats, the usage rate is low. When Meta wants to push Reels or ads, they find every possible way to get them in front of users. Do they really lack the ability to make E2EE the default for Instagram DMs? In the end, it’s not that users chose not to encrypt; it’s that Meta designed it so that users couldn’t encrypt conveniently. Offering the option in 2023 was just a token gesture to appease “privacy buffs.” Packaging the termination of E2EE as a “user choice” is pure PR.

“Defaults” are an incredibly powerful force, especially in internet products. Even if a setting can be changed with a single click, the extra operational and cognitive cost means that 90% of users will stick with the default. In the 90s, Netscape once had over 90% market share. Later, Microsoft made its browser the “default” by bundling Internet Explorer, flipping the entire market share. This triggered the largest antitrust investigation in US history. If Microsoft had explained in court that they only “defaulted” to Internet Explorer and users were free to switch to Netscape, the judge likely wouldn’t have bought it.

So, what is Meta’s true goal? I believe it is twofold: first, to satisfy the desires of various governments, and second (and more importantly), to obtain more data to train AI and for precision ad targeting. In an era where Large Language Models are extremely hungry for data, every private message is no longer just communication to Meta—it’s free “labeled data.” By terminating E2EE, Meta is essentially incorporating users’ private conversations into its “digital oil refinery.”

Anti-climax: If the product is free…

Having written over a thousand words, it might seem like I’m just here to bash Meta. But here is the anti-climax: to deal with governments and train AI, Meta’s termination of E2EE is actually understandable—one might even say perfectly reasonable. R&D, servers, and bandwidth all cost money. Yet, the moment a service charges a fee, users flee. If they don’t use your data to make money, are they supposed to run on “the power of love”?

Furthermore, Signal, and even Meta-owned WhatsApp and Messenger, all have E2EE as the default. The former even avoids storing user metadata and protects user privacy in every detail—and these tools are all free. If there are so many choices right in front of them and users still run to Instagram DMs, WeChat, or Telegram, clearly not caring about their own privacy, then what right do we have to blame a profit-driven tech enterprise for “creating value for shareholders”?

At the end of the day, most people only care about what is free, convenient, efficient, and has network effects. If privacy can even be a “fifth element,” that’s already not bad. This is the choice of the mainstream user. What “privacy buffs” can do is not to shout louder at tech companies, but to work harder on discourse and public education.

p.s. In the last issue, The price of privacy? HK$100k and 1 year in prison., I thought the topic was very important and the content was decent. As it turns out, the open rate was only 23.4%, the lowest in over a year. Does this happen to prove exactly what I was saying?

Last November, a severe fire occurred at Wang Fuk Court in Tai Po, claiming 168 lives and leaving nearly two thousand families displaced. Four months later, affected units have been repeatedly burgled, yet owners have not even had the chance to return home to collect their belongings, let alone receive arrangements for permanent housing. On the other hand, the humble wish of residents to hold an owners’ meeting has been delayed time and again. Basic facts—such as the distribution of the deceased and the status of the fire safety system at the time of the accident—were only first made public during a recent hearing. That hearing, however, lacked enough space for residents to attend, curiously lacked a live stream, and had no power to summon the contractors.

Despite this, Hong Kong citizens should perhaps be understanding. After all, the overall efficiency of the SAR government remains high. For instance, in safeguarding national security, nearly six years after the enactment of the Hong Kong National Security Law, the government again gazetted amendments to its implementation rules on Monday. These grant law enforcement agencies more power: if a designated person refuses to provide the password or decryption method for a seized electronic device, it constitutes a crime punishable by up to one year in prison and a fine of HK$100,000. For provisions with such deep impact on civil privacy, the government can issue them suddenly with immediate effect, bypassing Legislative Council scrutiny. This demonstrates a level of executive power that deserves commendation from national leaders.

As a Hong Kong citizen, I have no way to oppose such legislation other than “voting with my feet,” as hundreds of thousands who moved to other countries have done. But as a believer in and practitioner of the Cypherpunk spirit, I feel a minimal responsibility to raise public awareness. Since these provisions were announced and implemented so suddenly, citizens may not realize they are now in a jurisdiction where they can be ordered to unlock their phones at any time—where refusal is a crime facing a HK$100,000 fine and a year in prison.

In the digital world, to see is to possess

Some might think I am overreacting or even fearmongering. As I write this, it was only two weeks ago that Book Punch, prosecuted for the third time, was raided by law enforcement again. The founder, Pong Yat-ming, and three other staff members were taken away, this time accused of selling “seditious publications” such as the Biography of Jimmy Lai. Just one day after the new law was enacted, if law enforcement exercises their power to seize Yat-ming’s phone and demand it be unlocked, all his chat records—including his conversations with me—would be instantly exposed. This example happening right before our eyes clearly illustrates that privacy is a local and personal issue for everyone.

Some people like to say, “If you have done nothing wrong, you need not fear a knock at the door at midnight” (perhaps because a knock at dawn is more terrifying?). They say, “If you’re honorable and upright, why fear being seen?” I wonder if the gates of their homes are transparent, or if they walk around naked in the summer. These “upright” people don’t seem to understand that wearing pants doesn’t mean you’re afraid of someone seeing a gang tattoo on your buttocks. I am upright, and I don’t even plan to delete my chats with Book Punch, but I do not want people reading my records—simply because privacy is a fundamental human right.

Furthermore, the argument that the “upright have nothing to fear” completely ignores the difference between the physical and digital worlds. In the physical world, “seeing” and “possessing” are separated. A jewelry store allows you to see gold ornaments while maintaining security; unless you pay, you cannot possess them. But with data in the digital world, to see is to possess. Saying “why fear being seen” is like saying “why fear letting people enter your home and take whatever they want”—it is utterly absurd.

Imagine if I said: “Even if I see your creations, your reports, your messages, your photos, or your videos, I won’t take them, and I promise not to back them up.” That would be a rogue’s promise, because firstly, seeing the data is largely the same as possessing it, and secondly, you have no way of knowing if I’ve made a backup. Moreover, the new implementation rules for the National Security Law apply when the phone or electronic device has already been seized and is no longer in the owner’s hands.

If this is true for general data, privacy is even more critical for asset management. If a password is leaked, data can be copied; if you get your device back or log in from another device, you still have a copy. But digital assets are like physical objects; once stolen, the owner loses them permanently. Users of digital banking, centralized exchanges, and other custodial services already know the consequences of a password leak. The government itself frequently advises citizens to be alert to scams and never leak their passwords.

Providing a password = giving up all data and assets

I must supplement what disclosing a password means for self-custodied cryptocurrencies.

Last month, the South Korean government accidentally published the private keys of seized cold wallets online. Cryptocurrencies worth 6.4 billion won (about US$5 million) were stolen instantly. If you think this is just a one-off blunder, you are wrong; similar incidents have happened three times this year alone, involving amounts up to US$20 million. Ultimately, this isn’t just about being careful; it involves the workflow of law enforcement agencies and the core design of cryptocurrency and cryptography.

For most people today, a phone is not just for communication; it is a device that stores all passwords, 2FA, and passkeys used to verify identity. Having the phone and the password means being able to use all services, read all data, and move all assets as the owner. If “e-money” in a bank is stolen, at least there is a trail and the possibility of a freeze. Once self-custodied cryptocurrency is stolen, however, it will likely be sent to untraceable wallets, making it difficult to even identify the thief. The victim is left with no recourse.

I absolutely agree with the importance of protecting national security, but at the same time, private property is the cornerstone of society. Article 6 of the Basic Law states: “The Hong Kong Special Administrative Region shall protect the right of private ownership of property in accordance with law.” I do not intend to question the loyalty or bravery of the Hong Kong Police Force, but there are always “rotten apples.” For example, in February 2025, a male officer was involved in 12 counts of unauthorized access to police investigation systems (ESCC 553/2026); in early 2018, an exhibits officer allegedly stole approximately HK$860,000 from an exhibit room (FLCC 2551/2018); in May 2018, an officer stole HK$3,000 from a citizen’s wallet (ESCC 1273/2018; the list goes on). During an investigation, if just one person acts on greed, the owner’s property will vanish.

Even if a citizen is lucky enough not to encounter a dishonest officer, a simple mistake by an administrator—or even just following an outdated standard operation procedure, such as photographing evidence for records—is enough to cause a citizen who surrendered their password to lose their property. It is like the situation at Wang Fuk Court: the government prohibits residents from returning home to collect belongings “for their safety,” while allowing law enforcement and workers to enter, resulting in the theft of residents’ property.

The new rules of the Hong Kong National Security Law require citizens to provide passwords to law enforcement, but I see no accompanying measures to protect the property of affected citizens—specifically to ensure that self-custodied cryptocurrency does not vanish into thin air during the investigation.

I am not a legal professional; I can barely even parse the difficult, pieced-together Chinese sentences in the implementation rules. However, based on common sense and logic, if someone refuses to hand over a password to prevent their data and assets from being stolen, I believe that is perfectly reasonable.

If Hong Kong still had a normally functioning Legislative Council with representatives speaking for the people, they could at least voice these concerns. If there had been a public consultation as in a democratic society before the new rules were released, Cypherpunks—or even tech enthusiasts like me who know just a little—could have in good faith pointed out these potential loopholes.

Unfortunately, the new National Security Law rules are like the mandatory seatbelt law from months ago—rushed through without public consultation. Who can say for sure that there won’t be a department that accidentally gets something wrong, just like what happened with the seatbelt law? If a citizen hands over a password and their life savings are stolen, who will be held responsible?

p.s. While the new rules have drawn attention and many citizens seem to care about privacy, Meta’s announcement that it would terminate end-to-end encryption for Instagram DMs starting in May didn’t even cause a ripple. When there’s a chance, let’s discuss how Mark Zuckerberg ate his word and, more fundamentally, why end-to-end encryption is so vital.

3ook.com

Read, Listen, Own. The 3rd-gen bookstore ecosystem where authors and readers co-create a decentralized home for books and periodicals. By merging AI-enhanced experiences with Web3 ownership, we safeguard the freedom to read and the integrity of our shared history.

If I were limited to fifty words, the above would be my introduction to 3ook.com. However, since our readers seek deeper exploration, I will expand on the topic, focusing this time on the “Read & Listen” and “Utilizing AI” aspects.

Not Audiobooks, but Books with Voice

The most significant feature of 3ook.com is that it does not distinguish between text books and audiobooks. Every book can be switched between reading and listening modes at any time, unless the author chooses to disable the narration feature.

Although the setting of “Integrated Reading and Listening” may seem minor, it fundamentally changes how people interact with books, making the experience more logical and better suited to modern life. Taking my own experience as an example, among the five books I recently “read”—我城零五,時間的試煉, 浮世薔薇, Ｈ醫生一千零一夜,豆腐媽媽—I primarily listened to them while commuting, walking, hiking, or lying down, occasionally switching to reading mode as a supplement. Had it not been for this, even though I wanted to read these works, it would have been difficult to find the time and visual “HP” (health points) to finish them given my daily rhythm and eyes fatigued by long hours in front of a screen—not to mention the large volume of other excellent works available.

You might say that listening to books is nothing new, given that Audible has a twenty-year history. Indeed, I occasionally listen to audiobooks on platforms like Audible or Kobo. However, with traditional audiobooks, unless you purchase the text version separately, you cannot switch to reading mode at any time to clarify unfamiliar words, appreciate beautiful illustrations, or view infographics.

Furthermore, the vast majority of books simply do not have an audio version. Taking Amazon as an example, there are a hundred million text books, while its subsidiary Audible has only one or two million audiobooks—a coverage rate of less than 2%. Kobo fares slightly better; among nearly ten million books, about 5% offer an audio version. This means that 95% of titles are left behind. The absence of an audio version is the norm for e-books.

If you are reading popular titles like Atomic Habits or Chip War, finding an audiobook is not a problem. But if you are like me and interested in relatively niche subjects—especially Hong Kong books—the chance of finding an audio version is nearly zero. After all, even if these works spent their entire revenue, it might not be enough to record an audiobook using traditional methods.

Thanks to the rapid advancements in generative AI, 3ook.com pre-samples the voices of voice actors, analyzes their vocal characteristics, and synthesizes the entire text. The results are already quite impressive. While there are occasional inaccuracies in pronunciation or unnatural pauses, to be fair, were it not for these minor flaws, it has reached a point where it is hard to distinguish between a human narrator and an AI-generated one.

Not to self-deprecate, but objectively speaking, traditional e-book platforms are unlikely to hire someone to narrate 區塊鏈社會學 or 財富自由主義. As a niche author, if I wanted an audio version of my work, I had to find a way myself. Now that they are on 3ook.com, my six works and the upcoming 進擊的自由 are finally able to provide both text and audio versions simultaneously.

Designing 3ook.com began with satisfying my own needs, but I believe there are many readers, authors, and cultural preservationists facing the same issues.

Narrating Niche Works in Niche Languages

Just like niche works, niche languages cannot generate sufficient commercial scale.

Whether we are willing to accept it or not, Cantonese is becoming increasingly niche. I have immense respect for podcasters who love Cantonese and narrate Hong Kong books in our mother tongue, but we cannot deny that for most works, the market scale for Cantonese audiobooks is insufficient to support the cost of narration. However, with voice generation technology, there is hope for this problem to be solved.

3ook.com currently offers three default voice talents: travel expert Pazu, and my two partners Phoebe and Aurora. The former two narrate in Cantonese, while Aurora narrates in Mandarin. To be honest, compared to Mandarin, the flaws in Cantonese narration are slightly more frequent, but it is already well above a passing level. Considering that AI models are still improving rapidly, with enough training data, the quality of Cantonese narration will inevitably continue to advance, correcting that final 1% of error.

I am a bit sorry that 3ook.com cannot yet provide a Taiwanese (Hokkien) voice talent, as our resources are very limited. If there are Taiwanese readers who can provide support, we will certainly add it as soon as possible. Although I do not understand Taiwanese, I believe the intimacy of receiving information in one’s mother tongue, and the sentiment of preserving a gradually marginalized culture, is a universal value.

In addition to the default voice talents, 3ook.com launched the “Private Voice Talent” feature earlier this month, allowing members to upload their own or authorized voices to narrate works on their bookshelf. In recent weeks, almost every time I meet a friend, I forcefully invite them to record. A minute later, their voice can be used for narration. Sometimes the voice is so similar it’s indistinguishable; other times the effect isn’t quite ideal yet. The team is using various methods to fine-tune it, and we are confident that the precision of voice cloning will improve further in the near future.

Just imagine: a parent telling stories to their child, or a loved one narrating a novel for you. That “goosebump-inducing” feeling is something only the people involved can truly experience.

I am not an “AI enthusiast” who tries every new model daily, but from beginning to end, I adhere to the principle of “Humanity as the Essence, Technology as the Tool.” Through voice generation technology, 3ook.com aims to give every book a voice, narrated by the most familiar voices in our mother tongue, to preserve culture and keep the emotion alive.

P.S. After a 12-year absence, Stefanie Sun returned to Hong Kong for a concert last week. I attended with the feeling of catching up with my little sister. Seeing a slightly fuller Stefanie looking energetic, I was very happy for this mother of two. During the request segment, Stefanie invited Hong Kong people in the audience to raise their hands and then picked one. As soon as the person spoke, it was rhotic Mandarin—a Chinese person visiting Hong Kong for the concert. The next request was the same, and the third and fourth as well. It was a constant reminder that I was in “Hong Kong, China.” It wasn’t just the few picked for requests; by my “ear-test,” the six people to my right and at least three to my left were all from the mainland. Most of what I heard on the way in and out was Mandarin. It is not just Cantonese that is becoming increasingly niche.

This week, I’ve been even more persistent in my procrastination. By Wednesday morning, I still hadn’t put pen to paper; instead, I headed to the Kowloon City Magistrates’ Courts to observe the case against Book Punch and its founder, Pong Yat-ming, who are accused of operating an unregistered school. The initial pace of the trial was slow. Distracted, I was still mulling over a topic for this week’s newsletter—listening while thinking, thinking while listening. I realized that while the case has nothing to do with Web3, it is deeply connected to the freedom of information and the freedom of association. I decided to try something new: introducing the defendant, Book Punch, and recording the proceedings of this trial.

As this is a criminal case still in progress, I will only discuss the background information and the trial proceedings without offering any personal commentary. I will also suspend the opinion poll for one issue.

One Punch to Wake the Reader

Book Punch is an independent bookstore in Hong Kong. Founded in 2019, it is located on the third floor of an old building at 169 Tai Nan Street. The bookstore’s mission is “One punch to wake the reader,” a name and location that intentionally weave together reading, kung fu, and the Sham Shui Po neighborhood. Interestingly, the “Hong Kong and Kowloon Restaurant and Hotel Workers General Union,” where Ip Man lived and taught martial arts, is just dozens of meters away at 159 Tai Nan Street. This vicinity is also where Ip Man and Gong Er reunited and walked at night in the film The Grandmaster—the very streets where one “sees oneself, sees the world, and sees all living beings.”

The allegation of operating an unregistered school is not Pong Yat-ming’s first criminal case. In 2022, while hosting a reading group for Osamu Dazai’s Blue Bamboo (Memories of Sake), Book Punch provided three bottles of sake for participants to share. Two undercover officers who had signed up for the event arrested him, and he was later charged with violating the Dutiable Commodities (Liquor) Regulations, commonly known as selling liquor without a license. Additionally, because there were no “LeaveHomeSafe” posters or QR code requirements at the venue, he was charged with violating social distancing group gathering limits. Pong pleaded guilty to both charges and was fined HK$12,000.

Is a Language Class an “Education Course”?

Unlike the liquor license case, Pong has chosen to plead not guilty this time. Shortly after the session began, the Magistrate noted that under common law principles, the prosecution must prove the defendant’s guilt; the defendant is not required to prove their innocence.

Pong, who knows some Spanish, explained that he first started learning the language after discovering that the Spanish phrase for giving birth, “dar a luz,” literally means “to bring to light”—a concept that deeply moved him. Furthermore, World Book Day originates from a tradition in Catalonia, Spain, where men give flowers to women and women give books to men on St. George’s Day. Consequently, he organized a Spanish language and culture class last April to help more Hongkongers engage with the country’s language and culture.

The Spanish course was intended to run for 14 sessions. However, on April 23, which was exactly St. George’s Day, five inspectors from the Education Bureau arrived, declared the activity illegal, and collected evidence on-site. Subsequently, Book Punch and Pong Yat-ming were hit with five charges, including “managing an unregistered school” and “permitting a non-registered teacher to teach in a school.”

In court, Pong and his defense counsel—Lawrence Lau, the only defendant acquitted in the “47-person primary election” subversion case whose acquittal was upheld after a government appeal—agreed on basic facts: the floor plan of Book Punch, that a Spanish class was in progress, that there were 12 participants, and that the speaker, Montane Antonio Baro, was not a permitted teacher. However, they disputed the nature of the activity, arguing it was a “hobby class” rather than an “education course.”

Pong stated that before organizing the event, he researched relevant laws online. He found a 2017 response from the then-Acting Secretary for Education, Kevin Yeung, to a query from then-Legislator Shiu Ka-chun, stating that hobby classes do not fall under the scope of regulation and do not require registration.

The Education Bureau considers the purpose, content, and operation mode of a course to determine if it constitutes an “education course” under the Education Ordinance. Hobby classes such as painting, dance, and drama, as well as courses purely teaching skills or techniques, are not regulated by the Education Ordinance and do not need to register as schools if they do not provide educational activities.

— news.gov.hk (Hong Kong Government News)

During his testimony, Pong further argued that the Spanish class had no syllabus, no homework, no tests, no exams, and no certificates. It was “just for fun,” a relaxed way for participants to appreciate Spanish culture. Therefore, he viewed it as a hobby class, and according to both Kevin Yeung’s statement and public understanding, hobby classes do not require a license. Barrister Lawrence Lau argued that the defendant held an “honest and reasonable belief” that organizing the Spanish class was legal.

Defining the Law

In the closing arguments, Lau cited the Education Ordinances of 1913, 1952, and 1971. He pointed out that the core of the dispute—the concept of an “education course”—first appeared in 1952. Since there are no legal precedents regarding the definition of an “education course,” the court should interpret it based on legislative intent.

Lau further stated that while the defense does not dispute that the activity had “educational” elements, they deny it was a “course.” He cited definitions of “course” (課程) from various Chinese dictionaries and the Oxford English Dictionary:

A series of lectures, lessons, or instructional sessions in a particular subject; (chiefly British) a prescribed or planned curriculum or program of instruction in an academic subject, a vocational or practical skill, etc., esp. leading to an examination or qualification.

— Oxford English Dictionary

Lau argued that as the definitions in various Chinese dictionaries offered little reference value, and since in criminal cases where a term has multiple interpretations, the one most favorable to the defense should prevail, he suggested the court adopt the Oxford English Dictionary’s definition. Under this interpretation, the Spanish class—lacking a syllabus, homework, tests, exams, or certificates—clearly did not constitute an “education course” and should not be regulated by the Education Ordinance.

After hearing the testimony and arguments, Magistrate Lam Hei-wei adjourned for twenty minutes before announcing that the verdict would be delivered on April 10.

An Endless Cycle

Though a verdict is expected in a month, it doesn’t mean the matter is settled. Setting aside the fact that both parties have the right to appeal the April 10 decision, even if this case concludes, Pong and Book Punch have already been charged again—this time for violating the Places of Public Entertainment Ordinance. This stems from a “Standup 101 Graduation Performance” held at the bookstore last year following a stand-up comedy workshop. The bookstore and its founder pleaded not guilty last week, and the trial is scheduled for May 5.

Pong Yat-ming wants to stop and simply read in peace, but that seems increasingly difficult.

p.s. After the trial, I went with Pong Yat-ming and Gigi from Book Punch to Wanland Mori, the cafe featured in the cover photo. The lunch was great, the environment is pet-friendly with outdoor seating, and most importantly, it wasn’t crowded. I ended up staying there to work on this article until they closed at 6:00 PM. It didn’t feel like Hong Kong at all; it felt like Taiwan—that long-lost feeling of being able to spend a whole afternoon relaxing in a cafe.

Besides the Red Packets distributed through the DHK Newsletter (sponsored by Fluidkey), I personally put up $100 USD under the name of Just Books on Threads. I shared a link inviting 100 muggles to claim a $1 Red Packet as their first step into the Web3 world, hoping to bring more people on-chain during the Lunar New Year.

Goodwill Met with a Sybil Attack

Unlike the DHK Newsletter readership, the audience of Just Books has little to do with cryptocurrency. Furthermore, suspicious behavior like “giving away money” is often suppressed by social media algorithms. Initial engagement was slow: twenty some people claimed packets on the first day, only six on the second, and just one on the third.

I am naturally very “i”, and I wasn’t obsessed with forcing these packets out, so I didn’t push it further. However, around noon on the fourth day, a claim began appearing steadily every 2–3 minutes until the entire $100 was gone. This rhythm—not too many, not too few, no long gaps, but never twice in the same minute—had a probability of occurring naturally that was near zero. It was almost certainly the work of one person: Bob.

To use my analogy from last week about handing out $1 bills on the street: most passersby ignored me, but one person took a packet, walked around the corner, “passed by” again to take another, and repeated this until the bag was empty.

If you call me stupid, I won’t deny it. From the moment I pulled $100 out of my own pocket—or even eight years ago when I started sending crypto Red Packets—I was destined to be the fool. Civic education should be the government’s responsibility; why am I, a person of modest means, writing articles for free and “resorting” to paying out of pocket to get people to join?

But while I may be stupid, I wasn’t so naive as to not expect a Bob to perform a Sybil attack to “farm airdrops”. This is common in Web3. Before sending the first batch, I asked Fluidkey if there was a mechanism to prevent one person from claiming multiple times. They said yes, but I didn’t take it too seriously—not because I didn’t trust them, but because I knew such mechanisms only stop “honest people,” not “villains.” Fluidkey is positioned as a privacy wallet; if they push too hard by requiring phone verification to prevent Sybil attacks, it wouldn’t just be an inconvenience—it would be putting the cart before the horse and sacrificing privacy. Therefore, I don’t blame Fluidkey for failing to block Bob.

If Fluidkey has room for improvement, it’s in lowering the minimum amount per packet to ten cents or less. I originally wanted to set random amounts averaging much less than $1. The goal was to bring people on-chain; participants usually aren’t doing it for the money, but for the “good luck” and the fun of a game where everyone gets a piece. Converting someone from 0 to 0.1—or even 0.01—is an infinite times more. A smaller amount wouldn’t deter regular users, but it would significantly lower the economic incentive for “farmers.” Unfortunately, Fluidkey’s minimum is $1, so I had to spend $100 for 100 packets, which attracted the Sybil attack.

A Feature in Tech, a Bug in Application

Now, let’s talk about how the Fluidkey privacy wallet works. It’s too easy to get an AI to explain concepts these days, so I’ll just keep it brief.

Contrary to the public perception that crypto ensures privacy, “traditional” crypto transfers are like two people talking on Twitter: the transaction records are wide open for the world to see. For example, if DHK posts a wallet address for donations, anyone can search it on Etherscan to see the balance and history. When Alice sends money to DHK, she exposes her own address and her entire transaction history.

Fluidkey’s “magic” allows users to easily generate and manage infinite wallet addresses (using ERC-5564). Through an ENS address like ckxpress.fkey.id, the sender receives a brand-new stealth address every time they open it. One cannot check the history or link it to my other wallets unless I manually merge the assets.

Furthermore, through Account Abstraction (ERC-4337), Fluidkey uses smart contract wallets. Users don’t need to manage seed phrases (they use Passkeys) and can pay gas fees with assets other than ETH, like USDC. This makes transfers easier and enhances privacy; if you had to send ETH to every stealth address just to pay gas, the source of that ETH would eventually link all the addresses together.

By utilizing Account Abstraction and Stealth Addresses, Fluidkey solved three of the four major hurdles of giving away money. They even adapted to local customs by developing a Red Packet feature. However, Fluidkey’s core identity is a privacy wallet, whereas Red Packets are a social behavior. The tension between privacy and social interaction inevitably makes the group-distribution experience feel awkward.

In contrast, WeChat—which pioneered digital Red Packets—is a social app first, a wallet second. In a WeChat group, everyone sees who sent the packet, who claimed it, and who got the “Best Luck.” The money is just the packaging; the interaction is the point. In a privacy-centric wallet, all these interactions are (rightfully) discarded.

Fluidkey, which allows users to use multiple devices and emails to create multiple accounts—each generating multiple stealth addresses—is doing its job perfectly. But in the context of Red Packets, it is destined to never be as successful as a social app. This isn’t a “bug” that fails to identify users; it’s a “feature” that successfully protects privacy.

As for Bob: this giveaway relied on an honor system. It wasn’t even a hack. The money was already spent, so it wasn’t a financial loss to me. However, Bob deprived others of the chance to learn and enter the Web3 gateway. Furthermore, spending 150 minutes to net $70 means an “hourly wage” of only $28. He now has 70 wallet addresses to manage; he either has to make 70 separate small purchases or spend more time consolidating them, which risks being tracked on-chain anyway. Why be so exhausted? Why not leave the opportunity for the muggles, or write an article/video sharing the “farming” experience to help the community? That would be a much more meaningful contribution.

I have been doing this for eight years. I can tell you with full responsibility that the reality is: losing the money is hard, but giving it away is even harder.

Imagine this: if I pulled out 100 one-dollar bills on the street and tried to hand one to every passerby, what do you think would happen? I bet a “barrier” would form within a two-meter radius around me, and people would avoid me like the plague. Giving away money is, surprisingly, harder than handing out worthless flyers.

If “real money” from the U.S. is treated this way, distributing cryptocurrency is a hundred times more difficult. Last week, I used Fluidkey to send out 188 Red Packets worth a total of 228 USD. They were claimed almost instantly. While this might seem to contradict my point, it was actually the result of a massive cumulative effort: from the evolution of Bitcoin to Ethereum, over a dozen network upgrades, the emergence of stablecoins, the ongoing battle with the traditional system, the development of Fluidkey, and the hundreds of educational articles published by DHK Newsletters over the years. Without any single one of these links, we could never have achieved the smooth distribution we saw last week.

In addition to the Red Packets sponsored by Fluidkey for DHK Newsletter, I also funded several others out of my own pocket. Among them, one of the four “Opening Day” packets for Just Books went unclaimed by a volunteer. Furthermore, Just Books distributed 100 USD to the public; although it was eventually claimed after a few days, there is a story behind that which I’ll save for another time.

I also distributed Red Packets to other groups to promote cryptocurrency. In one group filled with scholars, journalists, and writers, I sent out ten packets and provided background info to “incite” participation. Only one person claimed it. Though this was expected, it reminded me of my essay in Moneyverse titled “When it Comes to Money, Everyone is pro-establishment.” Feel free to read the book for more; I won’t repeat it here, but the title says it all. Even public intellectuals find it difficult to imagine money outside of the established system, let alone practice it.

To give away Red Packets via cryptocurrency, at least four major hurdles must be overcome:

1. The Consensus of Money

This is the most basic, yet most overlooked point.

You might argue that the example of giving away 1 USD at the start of this article is only hard because the amount is too low. In reality, even giving away 100 USD per person isn’t necessarily easy. When the amount is low, people find it bothersome; when it’s high, they suspect a scam—thinking it’s too good to be true.

The most extreme and interesting example is from 2010, when Bitcoin core developer Gavin Andresen created the “Bitcoin Faucet,” giving away 5 BTC for simply completing a CAPTCHA. If that service existed today (though 5 BTC should be called a “Bitcoin Flood” rather than a faucet), the budget would be drained instantly regardless of size. Back then, however, it took a year and a half to give away 19,715 BTC—averaging only seven to eight claims per day. The reason was simple: the world didn’t believe what you were giving away was “money.” They either ignored you, labeled you a scammer, or thought you were a lunatic. Most people simply wouldn’t take it.

From Bitcoin to Ether, and from Ether to stablecoins, getting people to recognize that what is being sent is “money”—or an asset with strong “moneyness”—was not an overnight achievement. This single step took the blockchain community sixteen years.

2. Wallet Management

Even if society agrees that cryptocurrency is money, the public needs to own a wallet or be able to open one easily to receive it. This is the second threshold.

I’ve worked on LikeCoin for eight years. All these years, I’ve spent far more time educating people on how to open and manage wallets than I have discussing the philosophy of LikeCoin or decentralized publishing itself. It’s frustrating, but unavoidable. Unless you treat crypto solely as an investment tool and are happy with custodial services, you must master wallet management to achieve true self-sovereignty over your assets and data. There are no shortcuts here.

Fortunately, in the past year, embedded wallets and Passkeys have matured, making wallet creation much simpler. With the gradual implementation of “Account Abstraction” on Ethereum, I believe we will soon say goodbye to the era of “mnemonic phrases.” Managing a crypto account will feel almost identical to managing a Web2 service account.

The irony is that while registering for Web3 services like Fluidkey, Ether.Fi, or 3ook.com has become as easy as traditional services, I’ve found it harder to make users understand that—even if the experience is similar—self-management is fundamentally different from a custodial service. Perhaps this isn’t a problem; maybe users don’t need to know. Perhaps I’m just confusing my roles as an educator and a product designer. I don’t have the answer yet.

3. Payment of Gas Fees

Solving the first two problems only gets the money into their hands. How the recipient spends that money is another issue.

Unless you are sending Bitcoin or Ether directly, sending an ERC-20 token on Ethereum presents a problem: while receiving is free, the recipient cannot move the tokens without having ETH to pay for gas fees. Without ETH, the gift is useless—it’s just a digital souvenir.

The first time I widely distributed crypto Red Packets was in mid-February 2018. LikeCoin had just been minted on Ethereum earlier that month. To coincide with the Lunar New Year, we faced these two hurdles by sending out large numbers of “8 LIKE” packets. Those interested in “archaeology” can still find the old tutorials from that year.

Regarding the gas fee issue, we developed “Ethereum Token Payment Delegation,” which we shared at Open Source Developer Meetups. This technology later evolved into “sponsored transactions” and “paymasters.” This is why today I can send out USDC packets, and the recipient can spend that USDC by using the USDC itself for fees, without needing any ETH.

However, less than six months after we “solved” the gas fee problem, Ethereum became extremely congested, and fees skyrocketed. Since “sponsored transactions” meant we were footing the bill, we simply didn’t have the capital to sustain it. We were forced to stop the sponsorship and accelerate the development of the LikeCoin chain. That led to the story of LikeCoin v2.

Fun fact: To this day, there are still 6,726 wallets on the Ethereum mainnet holding 8 LIKE. They are most likely recipients of Red Packets in 2018. It shows just how much effort we put into promoting self-managed wallets, and how much of that effort ultimately resulted in “useless” digital artifacts.

4. Privacy for Both Parties

I have often emphasized that while cryptocurrency is very secure, it is not private. Even if you solve all the above issues, if the distribution mechanism isn’t handled correctly, the whole world can easily track the entire transaction history of both the sender and receiver. The simplest solution is to just not treat it as a problem—and honestly, that has been most people’s “solution” until now.

That is why I chose Fluidkey for my Red Packets this year. It wasn’t (just) that I was eyeing the sponsorship; it’s because it generates Stealth Addresses to decouple real identities, providing a level of privacy for both the sender and the receiver. Explaining the mechanics and usage of Fluidkey would require a full newsletter in itself, so I’ll leave it at that for now.

If you’ve read this far, I believe you’ll agree: giving away money is much more difficult than it looks.

The Year of the Snake, which just passed, could be described as a watershed moment for cryptocurrency gaining mainstream acceptance and becoming truly "usable." Many people began their crypto journey last year; even those who haven't yet jumped in have at least heard the news of governments, led by the US, gradually embracing Bitcoin, Ethereum, and stablecoins.

To me, the greatest significance of Web3 and cryptography lies in empowering citizens to autonomously choose, hold, trade, and even issue assets. Adding a million people to Bitcoin or Ethereum ETFs is far less meaningful than a single person opening a wallet and using cryptocurrency in their daily life. That is why, since entering this field in 2017, I have constantly looked for "excuses"—reasons—to give away money to encourage people to take that first step. These have included forwarding government handouts, Mid-Autumn Festival gifts, subscription rebates, and "Drifting Classroom" souvenirs. Among all these excuses, the most frequent and effective one remains the Lunar New Year Red Packet (Lai See).

As the saying goes, "an old trick is a good trick." Taking the opportunity of the Spring Festival, DHK Weekly is sending out 188 stablecoin red packets. The amounts are randomized, ranging from a minimum of 1.11 USDC to a maximum of 21.11 USDC, with a total value of 228 USD.

Unlike previous years where I often paid out of my own pocket, this year’s red packets are sponsored by Fluidkey. Thank you, Fluidkey. By clicking the link below, even readers who do not have Fluidkey installed can easily complete the process in just a few steps.

>> 188 USDC Red Packets, while supplies last

If you are one of the 188 readers who claimed a packet, you’ve likely experienced how seamless the process is. You might not be surprised—after all, WeChat pioneered digital red packets over a decade ago with similar ease. However, achieving that same level of smoothness in an open, diverse, decentralized ecosystem compared to a closed, centralized one is exponentially more difficult. This year, I only needed to set a few parameters—amount, quantity, and a greeting—to send them out, and readers only needed a few clicks to receive them. Furthermore, the fact that both parties cannot easily track each other's transaction history is not something to be taken for granted; it is the result of tireless efforts by countless developers, continuous community coordination, and years of cultivation.

I’d rather not make you read a long essay during the Lunar New Year. Let’s stop here for now. Later, we’ll chat about the sixteen-year journey of blockchain technology that made today’s seamless red packet experience possible.

P.S. Compared to "shopping" at malls, I prefer "marketing" at wet markets. Recently, while shopping for New Year goods at Fu Shin Street in Tai Po, I saw a dried seafood shop labeling conch as "Trump Conch." Playing dumb, I asked the owner why. The lady boss replied, "It’s from America, so it’s Trump!" Call it over-interpretation or just being sentimental, but I felt a bit gloomy hearing that. Our culture remains so accustomed to personality cults, often expecting a "Great Leader" to appear and welcoming individuals who override the system.

At the moment of writing, Jimmy Lai, the founder of Next Digital, had just been found guilty of three national security charges.

I still remember the end of last year; after attending the sentencing of the 47 pro-democracy activists, I found it impossible to sleep. Early the next morning, I simply got out of bed and headed back to the West Kowloon Magistrates’ Courts. In the wind and rain, I vied for a spot against the professional “line-standers” just to observe the trial of the Apple Daily national security case.

As soon as the court opened, the defense lawyer asked Jimmy Lai about his original intention for founding Next Digital. Mr. Lai replied that media promotes the flow of information; the more information one possesses, the more free one becomes. Thus, after the June 4th incident, he decided to found Next Magazine. He spoke with such nonchalance, yet hearing those words—unprepared as I was—made my skin crawl.

It wasn’t that the words were exceptionally moving. Quite the opposite: this discourse has become dated. If Mr. Lai hadn’t brought it up, I would have forgotten it entirely. Born in the era of the internet and smartphones, I witnessed information move from scarcity to expansion to total saturation. I can understand the context of that era. But for Gen Z, who have never “searched for data” and have only ever “swiped through posts,” these words might sound like “the more air there is, the smoother your breathing will be”—it doesn’t quite hold up to scrutiny. If more information made people freer, shouldn’t we have been free long ago? Why, then, do we still lack even universal suffrage?

Yet, Mr. Lai’s view at the time of the newspaper’s founding was absolutely correct. Information carries knowledge, facts, data, and the ability to discern right from wrong. Without information, one cannot fight for freedom, nor even know if freedom exists; fundamentally, the very concept of freedom cannot take root. Information freedom is the foundation of freedom of speech, personal freedom, freedom of belief, freedom of association, and all others. Information is like the forbidden fruit of Eden: one bite ensures you are no longer “domesticated,” liberating all subsequent possibilities while forcing you to bear a heavy price.

Thinking of this, the 1995 television advertisement for the launch of Apple Daily—the image of Mr. Lai hit by multiple arrows yet still calmly eating an apple—surfaced vividly in my mind, sending a shiver through me.

The Three Walls Obstructing Information

In the Japanese manga Attack on Titan, there are Nine Titans, each with distinct traits. The “Attack Titan,” hosted by the protagonist Eren Yeager, is characterized by its relentless pursuit of freedom throughout its life. The “Founding Titan,” also inherited by Eren, possesses the power to rewrite memories and control all other Titans (except the Attack Titan).

If we replace “Titans” with “Freedom,” then information freedom is the protagonist of this ensemble drama—playing the crucial role of “Attack on Freedom,” while also serving as the “Founding Freedom” from which all other layers of freedom derive.

By the same logic, the most direct way to restrict a citizen’s freedom is to block the flow of information. Coincidentally, much like the Maria, Rose, and Sina walls that kept the Titans at bay, the primary means of obstructing information also comes from three high walls.

from Attack on Freedom

Many friends know I am a massive fan of Attack on Titan. Even the title of my new work, Attack on Freedom, and its core imagery, are unabashed tributes to it.

Beyond Hajime Isayama’s brilliant plotting and profound characterization, fans of the series would likely agree that the dialogue is exceptionally well-crafted, embedded with many iconic lines. The one that left the deepest impression on me was the scene where the protagonist, Eren, mocks himself as a “Slave to Freedom.”

I’ve watched Attack on Titan over and over. I always understood this phrase—composed of a positive and a negative term—as being ultimately derogatory. Perhaps because I studied science and emphasize logic, I figured “positive times negative equals negative.”

Grammatically, “freedom” here is a noun. Just as some are slaves to money, Eren is a slave; it’s just that he pursues freedom instead of wealth. (And for the record, I absolutely detest Eren, though that’s beside the point).

On Monday, Jimmy Lai was sentenced to twenty years for conspiracy to collude with foreign forces and sedition under the National Security Law. Former executives Ryan Law, Lam Man-chung, and Fung Wai-kwong were also sentenced to between six years and three months and ten years. Though I was already pessimistic, the reality was still heartbreaking. That night, walking along the Shing Mun River and listening to podcasts, I thought about how these people have now literally become slaves to freedom.

As I walked, I suddenly realized that a “Slave to Freedom”, translated from 自由的奴隸, could have another, positive interpretation. The status of a “prisoner” is objective and imposed, but the attitude one takes toward misfortune is subjective—an embodiment of personal will. In this interpretation, 自由的奴隸 could be translated as a “Free Slave”.

Understood this way, 自由的奴隸, or Free Slaves, perfectly describes Jimmy Lai, Gwyneth Ho from the 47-person primary case, and Chow Hang-tung, who is still defending herself in her incitement to subversion case. Facing an absurd system, even knowing that a guilty plea or a plea for leniency could reduce their sentences—and in Mr. Lai’s case, knowing he didn’t even have to return to Hong Kong to face trial—they chose to face it bravely. They remained true to themselves from start to finish, upholding their beliefs and manifesting a noble character that pushes free will to its absolute limit.

I have no religious faith, but I believe that even if Jimmy Lai, now seventy-eight, were to pass away in prison, his free soul would live forever in heaven. However, the “heaven” I understand exists within the social conscience.

P.S. Whenever a gross social injustice occurs, I hesitate over whether to comment in this newsletter. The answer is often “no.” Sometimes it’s purely for the sake of “division of labor,” wanting to keep the newsletter focused on my main mission; sometimes it’s because, after stripping away the thoughts that would be illegal to write, there’s nothing left to say. And sometimes—like this time—it’s the belief that everyone already understands, and anything I write would just be repeating what is already known.

However, after being bombarded by the “News Simulcast” over the last few days—with phrases like “deserving of the crime,” “upholding the rule of law,” “justice served,” and even the classic line from From Beijing with Love, “traitor and running dog”—I made a last-minute decision. I cannot forfeit my right to freedom of expression. Even if what I write is merely common sense, in a society where common sense is scarce, it must be written—to add one more record to history.