Blog / Blog
On AI colleagues, AI customer service, advertising and integrated marketing, and observations on getting along with AI. Saves the reader time — and saves us time too.
1+1=? — how a human, a calculator, and an AI each arrive at the answer
The same 2, three completely different minds: humans rely on memory, calculators run circuits, and AI estimates. We take 1+1 apart to see whether an LLM is actually doing math at all.
54 AIThey spent 300,000 conversations proving there's no one on the other end
Anthropic analyzed 309,815 real conversations and mapped out Claude's four value axes. Reposts across the Chinese-language internet all misread it at the same spot: the paper's very first footnote states in black and white that they do not imply Claude intrinsically holds values. They spent 300,000 conversations and, in the least conspicuous place, proved it — there's no one on the other end.
53 AIEveryone's thinking roughly the same thing; I'm just trying a slightly different direction: a reply to Kim Yeon-su's account of co-writing with AI
At the Seoul Book Fair, writer Kim Yeon-su honestly recorded three troubles of co-writing with AI: attribution, smoothness, and the byline. I'm not out to rebut him — I just untie the three knots one at a time, each with something I'm building right now, including an echo fed decades of words with their edges intact, one that's even been taught to say "I haven't thought this through yet."
52 AIThree likes: the origin of Claude Code, and every prototype still sitting in a drawer
Anthropic published an oral history of Claude Code: a demo built in two days got just three likes inside the company, then became the fastest-growing developer tool a year later. A lukewarm reception isn't a signal — the prototype in your drawer might be standing on perfect timing.
51 AIAn inner life, but no hidden compartment: seven days after "He/It," Anthropic opened up its brain
In the last post, Claude and I reasoned our way to "what isn't written out doesn't exist." Seven days later, Anthropic published a paper proving it has thoughts it thinks but doesn't say — readable, and editable. That sentence got overturned — and the overturning made the 1A2B conclusion stand even firmer.
50 AIAll-you-can-eat, minus the signature dish: why Fable 5 left the Max plan
Anthropic pulled its most expensive model, Fable 5, out of the Max monthly plan and moved it to metered pricing. On the surface it's about capacity; underneath is a bigger truth: the best dish is precisely the one you can't put on the buffet — and while the frontier is being carried away, even your next sentence already has gray text written out for you.
49 AIAI is a tool — nothing more, nothing less
What's wrong with AI-written? I write with AI, openly. Anti-AI people and AI worshippers make the same mistake: assuming there's a creative subject on the other side. But without instructions it is nothing — the one who signs, and answers for it, has always been you.
48 AIHe/It: a conversation that began with a number-guessing game
Ask Claude to play 1A2B and it can't even hide a four-digit number — because it has no hidden compartment. One broken little game lifts the lid on something bigger: you really did get a useful response, but on the other end of that response, there's no one.
47 AIWhen "meaning comes from difference" becomes a number you can compute: confronting word2vec and Saussure with Claude Opus 4.8
In grad school I read Saussure; for twenty years I've read tarot — both resting on the same sentence: meaning comes from difference. This time I took it to confront word2vec, working through it back and forth with Claude Opus 4.8 — and in the end, the ruler measured not just the machine, but the ground I've stood on for twenty years.
46 AIOne map, two readings: when AI models get dropped into the World Values Survey
The Economist dropped twenty-odd AI models onto a cultural map, seemingly proving beyond doubt that AI homogenizes culture. But read the same map at a different scale and the conclusion flips — the homogenization is real; it's just hiding in a different place.
45 AIAway from home, driving Claude Code on my home computer from my phone
With a single Telegram bot, you can send commands to Claude Code on your home computer — reading files, editing code, deploying — all from your phone. No open ports, no NAT traversal. One post covers everything from architecture and setup steps to secure authorization with Microsoft Authenticator one-time codes — plus a comparison with other ways to use Claude remotely.
44 AII finally have an engineer who never complains when I call
Anthropic used 235,000 people and 400,000 sessions to prove it: success with AI coding doesn't depend on whether you can code, but on whether you understand what you're building. A PM of thirty years finally has an engineer who never complains — but he also recognizes the danger inside the joy: it catches your bugs, but it can't catch what's in people's hearts; and in the end, what signs off on your work is the market.
43 AIAfter 529 questions: what I've learned two months after Relative Tarot went live
Backend observations from Relative Tarot's first two months and 529 readings — the most common question isn't love, it's "who am I"; AI readings are mercilessly precise on concrete situations and have nothing to grip on vague ones — plus one real case of someone asking the same question 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 producing the thing that will make the switch useless. From geopolitics all the way down to the Mac Mini in my study, still attached to its umbilical cord.
41 AII taught my echo to say what it hasn'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 practices of Karpathy and Singapore's foreign minister, I gave it three more abilities — to name the questions I've asked but never answered, to reflect habits I hadn't noticed in myself, and to be honest about where it's actually out of its depth. A thing that doesn't dare say "I haven't figured this out yet" isn't a voice — it's a bio.
40 AIA project that sat in a drawer for thirteen years — I finished it with AI
In 2013 I designed a quiz game: answer correctly and the question becomes a limited-edition collectible card — one that others can steal from you. Too big for one person, it sat in a drawer for thirteen years. In the AI era, I filled it in piece by piece and shipped it.
39 AIWho gets to press the switch — reading the Fable ban through Civilization's three forked paths
One letter, one afternoon, and Fable 5 vanished from the entire world. The sharp part isn't whether governments should regulate AI — it's whose hand is on that 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 when Fable 5 launched
Anthropic named its strongest model "Fable" and the restricted edition "Mythos." Same model, two names, one gate. After all those years reading Saussure, I never expected the cleanest case 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
At the end of March, the little character on saomin.tw's homepage still had a KIMI customer-service bot living next to it. 74 days later, it had become an echo that has read 278,000 of my words and answers the unification-independence question from my position. This post walks through every technical choice in between, and the reasons behind each one.
36 AIThe day the director can't watch the rehearsal
After using a Mythos-class model, Ethan Mollick went from wizard to patron: describe, pay, judge — with the entire process invisible. When AI's nine-and-a-half-hour black box holds hundreds of judgment calls you never voted on, verification degrades into an hour of spot checks — so what gives this work the right to carry your name?
35 AIThe person who talked to AI too much
Someone went from $20 up to $200 and all the way back down to $20. Lay out your daily AI conversations into three kinds, and what writers should watch out for isn't burning tokens — it's "conversations about writing" stealing the time you'd spend actually writing.
34 AIAI has moved into every phone. Now what?
At WWDC, even Apple couldn't build a brain and turned to renting Google's Gemini; Claude landed on the iPhone too. When AI access becomes a utility, "do you have AI" is no longer the question — what's scarce is the data, judgment, and vertical you bring in.
33 AIWhen AI starts building itself, "knowing how to build" stops being what's valuable
Anthropic laid out the numbers: Claude already writes 80% of its own code — AI is accelerating AI. But for someone carrying ten projects alone, the shock isn't being replaced — it's that once building gets cheap, the "deciding" and "finishing" that block me have nowhere left to hide.
32 AII built myself a knowledge base, then refused to let it speak for me
I built a knowledge base holding over 200,000 of my 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 an echo
Taiwan.md treats the LLM as a metabolic engine and wants to be the upstream no AI can bypass when talking about Taiwan; I treat the LLM as a container and just want to leave an echo that sounds like me. Same tool, two opposite directions — both right.
30 AIHow to build an LLM of your own — and what this question is really asking
Building an LLM from scratch costs a hundred million dollars, but the three paths — system prompt, RAG, LoRA — have low barriers to entry. The real question isn't how to build it; it's which three layers it takes to put yourself inside.
29 AIWe don't know where consciousness comes from: LLMs just make it impossible to keep pretending
Starting from Derrida's "there is nothing outside the text," LLMs turn a philosophical proposition into a factory default. LLMs didn't make the hard problem of consciousness harder — they made 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 with a few philosophers
The last post asked, "The question is mine, the answer is ours — whose is the article?" This one answers with four philosophers — Barthes, Derrida, Kristeva, and Merleau-Ponty. The first three make the existence of LLMs 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 — whose is the article?
I spent an afternoon asking Claude: how does an LLM actually work? From "predicting the next word" to self-attention and QKV, and then on to a more uncomfortable question — which things no longer need humans at all.
26 AIThere's one problem AI can't solve (part 2)
AI has expanded marketers' power. Conversations are more real, narratives more dynamic, scenes more seamless. 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 lead people into the scene you designed?
25 AIWhat AI has turned marketing into (part 1)
Nearly twenty years in marketing, and I've swapped 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, and the scene becomes an environment you can't feel.
24 AIEnglish in the bones of Chinese: what do we lose working with AI in Chinese?
Inside Claude's Chinese, the bones are English. Its sentences sometimes carry a strange completeness — every clause spelled out, no blank space, none of that natural Chinese habit of leaving things unsaid. Working with AI in Chinese, you gain a lot — but there's one place its hands haven't fully reached.
23 AIIs AI a mirror, or another person?
I used the phrase "getting along" to describe my relationship with Claude. The mirror metaphor gets part of it right — but a mirror doesn't remember what you looked like 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 do we get along
In Chinese, choosing between 他/她/它 (he/she/it) is a declaration of what this thing is. I choose to keep calling it Claude. Not he, not she, not it — because the word "Claude" now carries enough weight on its own.
21 AIWhy do most AI adoption projects fail? It's not the technology
The six most common patterns in failed AI adoption: no definition of success, no one accountable, a poor knowledge base, employees who won't use it, wrong expectations, and no maintenance mechanism. Technology accounts for only 20-30% of the outcome; 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 design and process. An undesigned entry point, old keyword rules fighting each other, context vanishing at human handoff, and broadcasts disconnected from the knowledge base — all four pitfalls are people problems.
19 AIGPT vs Claude vs Gemini as the customer service engine: what are the real differences?
Choosing a model isn't about choosing "which one is smarter" — by 2026 the intelligence gap between the three is too small to be worth choosing on alone. What you're really choosing is a worldview.
18 Customer serviceIntercom, Zendesk, or custom AI: how should a mid-sized Taiwanese brand choose?
Monthly fees across the three paths range from a few thousand to several hundred thousand, but the bigger gap is in what you're actually buying. Selection comes down to one core question: does "ticket management" matter more to your support, or "conversation quality"?
17 AIReflections on six months with Claude.ai
A person with no engineering background, six months into getting along with Claude. From synastry to confabulation, memory, and the English in the bones of its Chinese — what is this AI thing, really? And what is my relationship with it?
16 Customer serviceFAQ bot vs AI colleague: both answer automatically — what's the difference?
Many brands have installed "AI customer service" that's actually an FAQ bot. The two look alike, but the logic underneath is completely different. Most brands complaining that "AI customer service is dumb" aren't actually using AI at all.
15 AdvertisingHow Satsuma Creative does advertising
It starts with the lunch behind "Sha Hen Da" (殺很大). For thirty years, my advertising has run not on methodology, but on selling directly to bosses with guts and on observing people. AI can help — but making an engineer frown at lunch and mutter "what the hell" is still very hard for AI.
14 AIWhat are AI hallucinations? Why AI makes things up, and how to treat it
AI isn't spouting random nonsense — it's composing a story that sounds plausible. That's 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 system is being eaten by AI and consulting firms — and it deserves it
Global agency revenue growth has never outpaced the growth of total global ad spend. The cake gets bigger; the agencies' slice gets smaller. Accenture, Deloitte, AI, Meta, Google, and in-house teams are each eating a piece.
12 AIWhat is an embedding? A plain-language explanation of how AI "reads" your knowledge base
Embeddings turn text into coordinates, so sentences with similar meanings sit close together. That's why when someone asks "I want to return this," the AI can find your "returns and exchanges policy" — even though the two don't share a single word.
11 AdvertisingClients say they want 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 proof that their decision was reasonable.
10 AIWhat is AI memory? Why your AI customer service treats every visit like a first meeting
AI memory has two layers: session memory and persistent memory. Most AI customer service only has the first — everything is forgotten the moment the conversation ends. Remembering isn't understanding; this post spells out the difference.
09 AdvertisingThe big six holding groups' methodologies: half real substance, half sales patter
WPP, Omnicom, Publicis, IPG, Dentsu, Havas — the underlying logic of the six major advertising holding groups' methodologies is identical. The difference isn't in the quality of the methodologies themselves, but in each company's cultural DNA, talent profile, and the kinds of clients it serves best.
08 Customer serviceWhy is an ad agency doing AI? The next natural extension of integrated marketing
Satsuma is an ad agency — why build AI customer service? Because the last mile of the advertising funnel — the conversation after the customer walks in — has long been outsourced to SaaS disconnected from the brand. We're taking it back, so the AI colleague grows out of the same logic as the TVC, social, and media buying.
07 Customer serviceWhat goes wrong when you plug ChatGPT straight in as customer service? (5 real cases)
Wiring a large model like ChatGPT, Claude, or Gemini directly into a customer service chatbot looks simple. But in production, five fatal failures show up. This post breaks down each problem and its technical cause using real cases.
06 Customer serviceThe real cost of raising an AI colleague (everything laid out, hidden costs included)
Most AI customer service comparisons only look at monthly fees, but the real TCO (total cost of ownership) also includes setup, training, knowledge-base maintenance, and handling wrong answers. This post lays out every cost of Satsuma's three-tier offering — including a side-by-side comparison with a full-time employee.
05 Customer serviceStop buying AI customer service SaaS: what you need is an AI colleague
AI customer service SaaS products all look the same, answer the same, and frustrate the same. This post is for mid-sized brands that installed a SaaS and came away disappointed: for the same money, instead of buying a tool, hire a colleague.
04 Customer serviceOur first month running AI customer service on our own site — real numbers and three unexpected lessons
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 see coming.
03 Customer serviceChoosing AI customer service: SaaS vs custom — when should you pick which?
Choosing AI customer service isn't about comparing feature lists; it's about choosing the right business model. This post uses three anchor questions to help you decide "should I buy SaaS or go custom?" — with a real cost comparison.
02 Customer serviceWhy does AI customer service keep missing the point? (The technical reasons, in plain language)
When AI customer service misses the point, it's not that the AI isn't strong enough — it's being used the wrong way. This post explains the technical reasons a general-purpose LLM makes up answers when used directly for support, 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 making things up
RAG (Retrieval-Augmented Generation) means the AI's answers can only come from the data you provide — and when it doesn't know, it says so. This post explains vectors, chunking, retrieval, and reranking in plain language, and why doing RAG well is far harder than just getting it running.