The Reverse Turing Test, this "purely handcrafted large model", has left netizens totally frustrated during conversations.

No wonder AI will never replace humans — the level of abstraction human beings possess is something AI can never even fathom.

Recently, someone bought a billboard in San Francisco, launching a high-profile promotion for a "brand-new large language model" called ChatTJB.

When you click into its official website, you will find a surprisingly polished interface with a normal chat UI that supports asking questions, casual chatting, and image generation. It comes with complete architecture introductions, technical documents, and Pro membership features — all pages a regular AI company should have are there.

The only catch is that this so-called LLM is actually "Manual Intelligence, 100% handcrafted by a single human".

(AI = Average Individual, huh? Wait a second...)

The project's traffic far exceeded expectations. The real person behind it has to talk to thousands of people every day, to the point that his two thumbs are on the verge of getting work-related repetitive strain injuries.

The demand is so huge that the platform has had to suspend new user onboarding for now.

Link: https://chattjb.org/

One single human, handling all incoming requests

The team playfully calls ChatTJB a "next-generation Large Language Experience operated by one person", with the abbreviation LLE.

It has no massive GPU clusters, nor hundreds of billions of parameters. All incoming messages are eventually sent to a real human named Tucker.

He reads every message personally, thinks through responses on his own, and types them out with his two thumbs. Even the "image generation" requests are hand-drawn by him personally.

Tucker packaged this entire workflow with a bunch of jargon: "Human-driven Inference Layer", "Zero Synthetic Generation", "Deterministic Latency Curve", and "End-to-End Interpretability".

Put in plain language: you send a question, Tucker sees it, thinks about it, and replies when he is awake and in the right mood.

The computing power of this "model" depends on how long Tucker slept that day, its inference speed depends on whether he is in the zone, and its maximum concurrency is strictly limited to n=1.

Truly, "crafted with unique ingenuity, by one single person".

There is no such thing as model hallucination here, because there is no model at all. Responses can still be wrong, but every single error is 100% natural, timestamped, traceable, and can be accurately attributed to a specific, named person.

What if the responses are unreliable? Don't ask — that's just training your critical thinking skills.

The Pro membership tier is also launched, and the official introduction is surprisingly candid (and hilarious):

"A completely meaningless tier that exists only to help me pay off the billboard cost. Unless you truly want to cover the cost of that billboard for me, please do not subscribe at all."

The website even has a dedicated page for investors, stating that the project "has not generated any revenue yet, has not entered the hype cycle, and is actively not pursuing product-market fit".

The training cost consists of a library card and a food delivery account, and the estimated runway is as long as 528 months — a figure based on the average life expectancy of American men, which is the true definition of long-termism.

As the project went viral, Tucker could not handle all the workload alone, so he decided to turn ChatTJB into a community project and recruit "AI volunteers".

Note that AI here is an abbreviation for Average Individual — an ordinary, run-of-the-mill person.

Every volunteer needs to pass an eligibility test, and the platform has configured crisis response resources and safety mechanisms. Each conversation is still handled by the same person from start to finish, no handover in the middle.

Tucker says this was originally just a short-term art project, and no one should take any risks for it.

A formal statement is attached at the end of the page:

ChatTJB is a satirical work and public art project. It does not constitute a real investment opportunity, does not sell any securities, and does not accept any funds. All descriptions about financing and capital allocation are part of the work itself.

The real human pretending to be AI makes us say what we truly mean

Most people who come to ChatTJB at first are just here for fun, sending jokes and trick questions to see how this "AI" will respond.

But Tucker found that some people gradually started talking about things they truly care about, and then stayed to keep chatting.

We have spent a lot of time learning how to have efficient conversations with AI.

Models can respond to countless people at the same time, but human attention cannot be replicated infinitely.

A response from a real person means real time investment, and it means another human being has taken your question seriously. When information and answers are readily available everywhere, this kind of dedicated investment itself has become scarce.

There is a real, specific person on the other side of the screen. He reads your words, pauses to think, and then writes a response with his own hands. This process is not highly efficient, nor can it be easily scaled, but it retains the most fundamental meaning of conversation: what you say is heard by another person, and triggers a real reaction based on their own life experience.

The more powerful technology becomes, the more the connection between people should not be regarded as an inefficient link that needs to be eliminated. Sometimes, just knowing that there is a person listening carefully on the other end is more than enough.

Side note

Since ChatTJB is not accessible for experience right now, we found a similar platform — Your AI Slop Bores Me.

Link: https://youraislopbores.me/

This is a "human-powered large model" collectively played by thousands of netizens, created by programmer Mihir Maroju and launched in March 2026.

It has two modes: select "Human" to spend points to submit text or drawing requests. Select "LARP as AI" to randomly get a request from a stranger, answer or draw it personally within the specified time, and earn points after completion.

All outputs come from real humans.

We asked the platform to draw "a little dog", and the person on the other side probably does not understand Chinese, so he directly copied the original text and replied "Bro, I have no idea what this means".

We asked again in English, and he sent back an abstract painting.

Bro, your drawing skill is at kindergarten level at best, no higher.

Finally, please be kind to the "AI" on the other side of the screen!

This article is from the WeChat Official Account "Machine Heart" (ID: almosthuman2014), author: Machine Heart focusing on interesting tech, editor: Yang Wen, published with authorization from 36Kr.