# A Thousand Sails Pass the Sunken Boat: A Freelance Engineer on the AI Content Factory
> A Chinese freelance engineer's YouTube channel takes a rare break from hardware to talk about its own industry: collapsing creator traffic. Short-form video is draining watch time while AI content factories let one person run ten channels; the platform's ability to detect AI content has an expiry date and is already wrongly banning humans. His conclusion: you aren't competing with AI, you're competing with people who wield it. His plan: use the three-to-five years in which being human is still provable. My extension: content can be mass-produced, liveness cannot; faces expire, records don't.
Published: 2026-08-14
Locale: en
Tags: lunchuizhe, ai-content, creator-economy, youtube
TL;DR: Creator traffic is collapsing under two stacked forces: short-form video and AI content factories. AI's real advantage isn't lower cost per video — it's replication: once a workflow works, the tenth channel is nearly free. Anything with near-zero marginal cost becomes a red ocean; scarcity migrates to what can't be copied. He's spending the three-to-five-year window in which humanity is still provable. My addition: a face expires, a record you dare to be graded on doesn't.
> *Humming the old flute tune to friends long gone,*
> *I come home a woodcutter whose axe-handle rotted while he watched one game.*
> *Past the sunken boat, a thousand sails go by;*
> *ahead of the sick tree, ten thousand trees are in spring.*
> —— Liu Yuxi, "Reply to Bai Juyi at Our First Meeting in Yangzhou" (Tang dynasty, 826; translation mine)
## What this video is about
Lunchuizhe (掄錘者, "the hammer swinger") is a Chinese freelance engineer with a YouTube channel of roughly ten thousand subscribers, normally devoted to local AI hardware and agent workflows — modded 48GB RTX 4090s, Hermes, DeepSeek, that crowd. This video (2026-08-11, 14:47) takes a rare break from hardware to talk about the industry he himself sits in: the collapse of Chinese-language creator traffic on YouTube.
The starting point is a creator he likes, "German Knowledge," announcing a pivot: channel income has fallen below €1,000 a month (the creator's own account, as of 2026-08), which no longer covers a decent life in Germany — though having bought a few houses early on, he can live off the rent, and at fifty he still wants one more swing at it, so he's off to Africa to film new material. Lunchuizhe is careful to note that he disagrees with this creator on politics almost completely and watches him anyway — he watches anyone who can sit in front of a camera and calmly explain a problem, whatever their view of China. That opening sets the register for the whole video: the subject is a livelihood, not a position.
## The notes I took
**The traffic collapse is two forces stacked, and AI is only one of them.** The first is short-form video: since TikTok, short-form has overtaken long-form in views and total watch time, and the gap keeps widening (his description; no figures given). The second is the AI content factory: in a niche where one person used to own the whole market, someone with a working AI pipeline can now run ten channels at once. His arithmetic: your share drops from 1 to 1/11 — and so does each of the factory's channels, but the factory has ten of them.
**The counterintuitive part: AI video isn't cheaper. It's copyable.** This is where he explicitly disagrees with German Knowledge. Talking into a camera and letting AI handle the edit is dirt cheap; generating fifteen or twenty minutes of video with models like LTX or Minimax H3 takes a dozen hours or more, ties up expensive GPUs, produces piles of discarded takes, and burns tokens. AI's real advantage is replication: once the workflow is dialed in, the marginal cost of the second channel — or the tenth — approaches zero.
**The whistleblower runs AI channels himself, and he refuses to condemn the business.** He's open about it: several of his English-language channels are AI end to end — he supplies the ideas and the final review; voiceover, imagery and editing are automated. But he deliberately keeps the scale small, for two reasons: multiple accounts can't survive platform enforcement risk, where one ban kills everything under the account; and more fundamentally, the barrier is too low — "whatever I can master, others can master too; this model can't last." As for the red ocean itself, he calls it healthy competition: the people who invested in these tools have a monetization window of a few months to a year, and if they don't mass-produce inside that window, how do they ever recover the up-front cost? He quotes the old line about wildfire never burning the grass out — the spring wind brings it back. One playbook dies, another grows.
**Platform detection has an expiry date, and it's already convicting the innocent.** Today YouTube can still catch AI content by diffusion-model artifacts in the image and voiceprint signatures in the TTS, and purge it wholesale. He was a wrongful conviction himself: a small channel he ran as a real human was flagged as AI garbage and deleted without warning, then restored months later, equally without warning and without a stated reason. His forecast: within two, three, at most five years, no algorithm will separate human from AI creators at a cost the platform can bear — that road gets closed by the technology itself.
**The most important line of the video: you aren't competing with AI. You're competing with people who wield it.** AI has no consciousness; it doesn't want money or traffic. Competition is always between humans. So avoiding AI isn't an option — if you don't use it, someone else will, and the shift in how content gets made will reach you either way.
**His survival plan is to spend the three-to-five years in which being human is still provable.** Starting now: an on-camera channel under his real identity, much more livestreaming (real-time reaction is the part he thinks AI will need a decade or more to match), an audience built to one or two hundred thousand, plus his own website and community across a matrix of platforms. By the day AI can no longer be told apart from humans, whether the face on screen is him or his digital twin won't matter — what matters is that the IP is already publicly established as run by a real person. Content brings the traffic; the community holds the trust.
## Follow-on thinking
**1. What he's really describing is a mismatch between fixed and marginal cost — the same structure as the OCP piece.**
The cost structure of AI content: getting the workflow to work is expensive (hours, GPUs, discarded takes, tokens); copying it afterwards is nearly free. Anything with near-zero marginal cost turns into a red ocean by arithmetic, not by ethics — which is exactly why "this model can't last" needs no moral argument. And scarcity always migrates to where marginal cost refuses to fall: real-time reaction, first-hand experience, a human willing to be accountable for a judgement.
In the OCP piece I left a line: the protocol can be opened, the brazed seam cannot. The content-industry version is: **content can be mass-produced, liveness cannot.**
**2. As proof of humanity, a face expires. A record doesn't.**
His window rests on the fact that showing your face still proves you're human. But by his own forecast, in three to five years neither face nor voice will be evidence. What's left?
My answer is **a verifiable record**: publish your judgements, lock the criteria, grade them when they come due. An AI content factory can't produce that — not because AI can't write judgements, but because the factory's strategy is hit-and-run; it will never volunteer a long history that can be held against it. And any account that does volunteer one is, behaviorally, indistinguishable from a person. This is what we practice here: 84 Chinese-language posts on this blog before this one (as of 2026-08-14), AI assisting with the material, the judgements and conclusions human, the wrong calls left standing to be graded. Faces expire, voices expire; a record you dare to be graded on doesn't.
**3. By the data, his own channel hasn't collapsed — which makes his actions more informative than his argument.**
I pulled view counts for his channel's twenty most recent videos (yt-dlp snapshot, as of 2026-08-14): most sit between 3,000 and 8,600 views, healthy engagement for a channel of about ten thousand subscribers; this video stood at 3,672 three days after upload, within the normal range. In other words, the alarm isn't coming from his dashboard — it's coming from watching a peer's. He is pivoting before his own numbers bleed: real identity, livestreams, community, the platform matrix, all positioned early.
That is more persuasive than any argument: **judge a person not by what they say, but by what they do before they have to.** One honest caveat: views accumulate with age, so newer videos naturally read low; this snapshot can say "no visible collapse," not "nothing is wrong."
## Further reading
- Original video: Lunchuizhe, "YouTube creator traffic is falling; AI batch factories plus short-form drain are squeezing creators — embracing AI is the real way out" (2026-08-11, 14:47, in Mandarin)
- His earlier companion piece on the same theme: "How to handle falling YouTube and website traffic? Make content for AI first, humans second"
- Channel view counts: public data pulled via yt-dlp, as of 2026-08-14; figures change over time
- The two couplets at the top are from Liu Yuxi's reply poem to Bai Juyi (826). The woodcutter is from the legend the poem leans on: he watched immortals play one game of chess, and when it ended his axe-handle had rotted and a century had passed at home. My footnote, not the show's.