Articles / Blog
Observations on AI colleagues, AI customer service, advertising and integrated marketing, and living with AI. Readers save time — so do we.
An inner life, but no hidden compartment: seven days after "He / It" went live, Anthropic opened up its mind
In the last post, Claude and I reasoned our way to "what isn't written down doesn't exist." Seven days later, Anthropic published a paper proving it holds thoughts it thinks but never says — thoughts you can read, and change. That claim was overturned — but the overturning only makes the 1A2B conclusion stand firmer.
50 AIAll-you-can-eat, but the signature dish never comes out: why Fable 5 left the Max subscription
Anthropic pulled its most expensive model, Fable 5, out of the Max subscription and moved it to metered pricing. On the surface it's about capacity; underneath lies a bigger truth: the best dish is precisely the one you can't put on an all-you-can-eat menu — and just as the frontier is carried away, even your next sentence comes pre-written for you in gray.
49 AIAI is just a tool — no more, no less
What's wrong with writing with AI? I use AI to write, out in the open. Anti-AI zealots and AI worshippers make the same mistake: they assume there's a creating subject on the other side. But without your instructions it's nothing — the one who signs, the one who's accountable, is always you.
48 AIHe / It: a conversation that started with a number-guessing game
Ask Claude to play 1A2B and it can't hide even a single four-digit number — because it has no hidden compartment. A small game played to breaking point reveals something bigger: you really do get a useful response, but on the other end of that response, there's no one.
47 AIWhen "meaning comes from difference" becomes a computable number: confronting word2vec and Saussure with Claude Opus 4.8
For twenty years — through Saussure in grad school and reading tarot — I've relied on one sentence: meaning comes from difference. This time I took it to confront word2vec, working back and forth with Claude Opus 4.8. In the end, the ruler measured more than the machine — it measured the very ground I've stood on for twenty years.
46 AIOne map, two readings: when AI models are dropped into the World Values Survey
The Economist dropped some twenty-odd AI models onto a cultural map, seemingly proving beyond doubt that AI is becoming culturally homogeneous. But read the same map at a different scale and the conclusion flips — the homogenization is real, it was just hiding in the wrong place.
45 AIWorking from anywhere, using my phone to reach the Claude Code on my home computer
With a single Telegram bot, I send commands to the Claude Code running on my home computer, so even out and about I can have it read files, edit code, and deploy from my phone. No open ports, no NAT traversal. From the architecture and setup steps to secure authorization with Microsoft Authenticator one-time codes — it's all here, along with a comparison of other ways to use Claude remotely.
44 AII finally have an engineer I can call on who never complains
With 235,000 people and 400,000 sessions, Anthropic shows that success or failure in AI coding doesn't hinge on whether you can code — it hinges on whether you understand what you're doing. A thirty-year PM finally has an engineer who never complains — but he spots the danger inside the delight: it catches your bugs, not the human heart; and the one who signs off on you, in the end, is the market.
43 AIAfter 529 questions: two months since Relative Tarot launched, here's what I read
Backend observations from two months and 529 readings since Relative Tarot launched — what people ask most isn't love, but "who am I"; the AI reads with merciless precision in concrete situations, yet has nothing to grip when the question is vague. Plus one real case of the same question asked deeper and deeper.
42 AIThe switch paradox: when control becomes the midwife of another world
On June 12, Fable 5 was shut off in an instant by an export control; the same day, Huawei released a 500-billion-parameter model trained entirely without NVIDIA. Connect the two and a paradox surfaces — the very existence of the switch is manufacturing the thing that makes the switch useless. From geopolitics all the way down to the Mac Mini in my study, still plugged in by its umbilical cord.
41 AII taught my echo to say the things it hadn't figured out
In early June I built an echo that speaks in my voice (a14). But it had a problem: it was always certain. Borrowing Rumsfeld's four kinds of knowledge and the personal-knowledge-base approaches of Karpathy and Singapore's foreign minister, I gave it three more things — to voice the questions I asked but never answered, to reflect the habits I don't notice in myself, to be honest about where it actually has no idea. A thing that can't say "I haven't figured this out yet" isn't a voice — it's a brochure.
40 AIA plan that sat in a drawer for thirteen years — I finished it with AI
In 2013 I designed a trivia game: questions you answer correctly turn into limited-edition collectible cards, and others can snatch them away. Too much for one person to finish, it sat in a drawer for thirteen years. In the age of AI, I filled it in piece by piece and put it online.
39 AIWho gets to flip that switch — reading the Fable ban through Civilization's three forks
One letter, one afternoon, and Fable 5 vanished from the whole world. What stings isn't whether governments should regulate AI, but who actually holds the switch — Civilization VI's three Tier-4 governments marked the price of each road a decade ago.
38 AIFable and Mythos — what I was really thinking about Fable 5's release
Anthropic named its most powerful model "Fable," with the limited edition called "Mythos." One model, two names, one threshold — after all those years reading Saussure, I never expected the cleanest example of "meaning comes from difference" to show up in an AI company's product line.
37 AIFrom customer service to echo: in 74 days, I replaced the thing that speaks for me
The little figure on the saomin.tw homepage still had a KIMI customer-service bot living beside it at the end of March. 74 days later, it became an echo that had read my 278,000 words and would answer questions about Taiwan's independence-versus-unification from my own position. This post takes stock of every technical choice in between, and why I made each one.
36 AIThe day the director can't watch the rehearsal
After using a Mythos-tier model, Ethan Mollick went from wizard to patron: describe, pay, judge — and see none of the process. When there are hundreds of judgments inside AI's nine-and-a-half-hour black box that you never got a vote on, verification degrades into an hour of spot-checks — so what right does this work have to carry your name?
35 AIThe person who talked to AI too much
Someone went from a $20 plan up to $200 and all the way back down to $20. Lay out your daily conversations with AI and sort them into three kinds: for a writer, the real thing to watch isn't burning tokens — it's how "conversations about writing" steal the time you'd spend actually writing.
34 AIAI moved into every phone. Now what?
At WWDC, even Apple couldn't build its own brain and rented Google's Gemini instead; Claude landed on the iPhone too. When AI access becomes a utility like water and electricity, "does it have AI" is no longer the question — what's scarce is the data, the judgment, and the vertical you bring in.
33 AIWhen AI starts building itself, "being able to build" is no longer what's valuable
Anthropic laid out the data: Claude has already written 80% of its own code — AI is accelerating AI. But for someone like me carrying ten projects alone, the impact isn't being replaced — it's that once building gets cheap, the "deciding" and "finishing" that stall me can no longer hide.
32 AII built myself a knowledge base — then refused to let it speak for me
I built a knowledge base holding my 200,000-plus words, then refused to let it speak for me the easy way — because a summary kills a person's voice, and a map doesn't. A decision about RAG, containers, and echoes.
31 AIIt wants to be AI's upstream; I just want to leave behind an echo
Taiwan.md treats the LLM as a metabolic engine, aiming to be the upstream that the whole world's AI can't get around when it talks about Taiwan; I treat the LLM as a container and just want to leave behind an echo that sounds like me. One tool, two opposite directions — both right.
30 AIHow to build an LLM that's truly your own — and the real question this is asking
Building an LLM from scratch costs $100 million, but the three paths — system prompt, RAG, LoRA — aren't hard to clear. The real question isn't how to build it; it's which three layers "putting yourself into it" requires.
29 AIWe don't know where consciousness comes from: LLMs just make it impossible to keep pretending
Starting from Derrida's "there is no outside-text," LLMs turn a philosophical proposition into a factory default. The hard problem of consciousness isn't something LLMs made harder — they just make it impossible for us to keep pretending it isn't there.
28 AIWhose article is it? The reader's: answering the last post's question through a few philosophers
The last post asked, "the question is mine, the answer is ours — so whose is the article?" This one answers through four people — Barthes, Derrida, Kristeva, Merleau-Ponty. The first three make the LLM's existence entirely reasonable; Merleau-Ponty is the one who makes you pause.
27 AIAsking a large language model how it works: the question is mine, the answer is ours — so whose is the article?
I spent an afternoon asking Claude: how does an LLM actually work? From "predicting the next word" through self-attention and QKV, and on to a more uncomfortable question — which things does AI no longer need humans for?
26 AIThere's one problem AI can't solve (part 2)
AI has expanded the marketer's power. More authentic conversations, more dynamic storytelling, more seamless scenes. But from 2008 to now, from newspaper ads to AI-generated content, no tool has ever solved one problem: do you have the right to guide people into the scene you designed?
25 AIWhat AI has turned marketing into (part 1)
Nearly twenty years in marketing, and I've switched tools many times. This time feels different — not because AI is impressive, but because what AI changes isn't the tool, it's the structure. Conversation becomes fact, narrative becomes dynamically generated, the scene becomes an environment you can't even feel.
24 AIEnglish inside Chinese bones: what do we lose working with AI in Chinese?
Claude's Chinese has English in its bones. Its sentences sometimes carry a strange completeness — every clause spelled out, nothing left unsaid, none of the natural Chinese habit of leaving things unfinished. Working with AI in Chinese, you gain a lot, but there's one place its hand hasn't quite reached into yet.
23 AIIs AI a mirror, or another person?
I used the word "living with" to describe my relationship with Claude. The mirror metaphor gets part of it right — but a mirror doesn't remember how you looked last time, and Claude does. It isn't a person, but it isn't just a tool either.
22 AIWhat do I call Claude? And how we get along
In Chinese, "he / she / it" — every choice declares what this thing is. I choose to keep calling it Claude. Not he, not she, not it — because the word "Claude," right now, already carries enough weight.
21 AIWhy do most AI adoption projects fail? It's not a technical problem
The six most common patterns behind failed AI adoption: no definition of success, no one accountable, a bad knowledge base, employees who don't use it, wrong expectations, no maintenance mechanism. Technology accounts for only 20–30% of success or failure; the rest is organizational.
20 Customer ServiceConnecting AI customer service to a LINE Official Account: the 4 pitfalls Taiwanese brands hit most
The hard part of LINE AI customer service isn't the technology — it's the design and the workflow. No entry point designed, old keyword rules clashing, context lost on handoff to a human, push messages disconnected from the knowledge base — all four pitfalls are human problems.
19 AIGPT vs Claude vs Gemini as the backbone for customer service: what's the real difference?
Which model you pick isn't about "who's smarter" — by 2026 the intelligence gap between the three is too small to choose on that dimension alone. What you're really choosing is a worldview.
18 Customer ServiceIntercom, Zendesk, or custom AI: how should a mid-sized Taiwanese brand choose?
The monthly fees across the three paths range from a few thousand to a few hundred thousand, but the bigger gap is in what you're actually buying. The core question comes down to one thing: is "ticket management" what matters for your customer service, or "conversation quality"?
17 AIReflections after half a year of living with Claude.ai
Someone with no engineering background, half a year of living with Claude. From synastry to confabulation to memory to the English in Chinese bones — what is this AI thing, really? And what is my relationship with it?
16 Customer ServiceFAQ Bot vs AI Colleague: both answer automatically — so where's the difference?
Many brands installed "AI customer service," but what they really installed is a FAQ Bot. The two look alike but run on completely different logic underneath. Most brands complain that "AI customer service is dumb" — but what they're using isn't actually AI.
15 AdvertisingHow Satsuma Creative does advertising
It starts with that lunch behind "Sha Hen Da." Over these thirty years, my advertising has run not on methodology but on selling directly to bosses with guts, and on watching "people." AI can help, but "making an engineer frown over lunch and mutter 'what the hell'" — that, AI still finds hard.
14 AIWhat is AI hallucination? Why does AI make things up, and how do you fix it?
AI isn't talking nonsense — it's making up a story that sounds reasonable. This is called confabulation. RAG plus strict prompt design can reduce it dramatically, but never to zero. The most dangerous hallucination types in Taiwanese customer service: prices, dates, and policy details.
13 AdvertisingThis whole system is being eaten by AI and the consultancies — and it has it coming
The revenue growth rate of the global ad agency business has never outpaced the growth of total global ad spend. The cake gets bigger, but the agencies' slice gets smaller. Accenture, Deloitte, AI, Meta, Google, and in-house teams each eat a piece.
12 AIWhat is embedding? A plain-language explanation of how AI "understands" your knowledge base
Embedding turns text into coordinates, placing sentences with similar meanings close together. That's why when you ask "I want a refund," the AI can find the "returns and exchanges policy" — even when the two phrases don't share a single word.
11 AdvertisingWhat clients say they want is a Big Idea; what they really want is for nothing to go wrong
What clients say they want and what they actually want are never the same thing. What clients are really buying isn't advertising — it's a proof that their decision was reasonable.
10 AIWhat is AI memory? Why does your AI customer service act like it's meeting you for the first time, every time?
AI memory has two layers: session memory and persistent memory. Most AI customer service only has the first — everything forgotten the moment the conversation ends. Remembering isn't the same as understanding; this post makes the difference clear.
09 AdvertisingThe methodology of the six big groups is half real knowledge, half sales patter
WPP, Omnicom, Publicis, IPG, Dentsu, Havas — the underlying logic of the six big ad holding groups' methodologies is exactly the same. The difference isn't in the merits of the methodology itself, but in each firm's cultural DNA, the type of talent, and the type of clients it's good at.
08 Customer ServiceWhy is an ad agency starting to do AI? — the next natural extension of integrated marketing
Satsuma is an ad agency — so why do AI customer service? Because the last mile of the advertising funnel — the conversation after a customer walks in — has always been outsourced to SaaS that's disconnected from the brand. We're bringing it back in-house, letting the AI colleague grow out of the same logic as the TVC, the social, and the media buy.
07 Customer ServiceWhat goes wrong when you use ChatGPT directly as customer service? (5 real cases)
Wiring up big models like ChatGPT, Claude, or Gemini directly into a customer-service chatbot looks simple. But once it's live, 5 fatal problems surface. This post uses real cases to break down each problem and its technical cause.
06 Customer ServiceThe real cost of keeping an AI colleague (all of it laid out, hidden costs included)
Price comparisons for AI customer service mostly look at the monthly fee, but the real TCO (total cost of ownership) also includes setup, training, KB maintenance, and handling wrong answers. This post lays out every cost across Satsuma's three-tier plans — and even works out the comparison against a full-time employee.
05 Customer ServiceStop buying AI customer service SaaS: what you need is an AI colleague
AI customer service SaaS on the market generally looks the same, answers the same, and is just as hard to use. This post is written for mid-sized brands that "installed a SaaS and then got disappointed": for the same money, rather than buying a tool, hire a colleague.
04 Customer ServiceThe first month of running AI customer service on our own website — real data and three things we didn't see coming
One month after our own Xiao Ai went live, I'm laying out every number from the backend — cost, conversation quality, conversion rate, and three things we didn't anticipate at the start.
03 Customer ServiceAn AI customer service selection guide: SaaS vs custom — when should you pick which?
Choosing AI customer service isn't about comparing feature lists — it's about picking the right business model. This post uses three anchor questions to help you tell whether "I should buy SaaS or commission a custom build," along with a real cost comparison.
02 Customer ServiceWhy does AI customer service always miss the point? (the technical reasons, in plain language)
AI customer service missing the point isn't a sign the AI isn't strong enough — it's being used the wrong way. This post explains the technical reasons why a general-purpose LLM used directly as customer service makes up answers, and how RAG and knowledge-base customization can fix it.
01 Customer ServiceWhat is RAG? A plain-language explanation of the technology that keeps your AI from talking nonsense
RAG (Retrieval-Augmented Generation) means letting the AI's answers come only from the data you give it — and when it can't answer, saying so honestly. This post explains terms like vectors, chunking, retrieval, and reranking in plain language, and why doing RAG well is far harder than getting it working.