25 comments

[ 0.14 ms ] story [ 30.1 ms ] thread
It's just another Qwen 3.6 27B finetune (not even 3.8...), claiming to beat TB-sized models. This is probably the 57th model like that, just this week, claiming the same thing... And they never deliver, to say the least.
Didn’t they deliver something here? Why does it matter what base model they used?
What they all deliver is benchmaxxed models that are awful in actual use.
Qwen is a really awesome model to use!
I'm giving out more unpopular opinions today for entertainment and thought :)

Every model is benchmaxxing today. Many of them are awful until tuned for any particular task.

The finetunes will continue until morale increases.
> (not even 3.8...)

3.6 is better for non-coding tasks, noticeably so.

The article appears to be slop and even has random model train of thought text interleaved with the article text at various places.

And you can also tell that this is an obviously Chinese model trained on the Chinese internet in the CCP alternate reality, due to various wild claims:

> Looks a bit unfamiliar, doesn't it? Don't worry, its founder and CEO is an old acquaintance: Chen Danyan.

> Known as the "godfather" of programmers in the internet era, his achievements are a matter of public record:

> Shanda Network's co-founder and the head of Lian Shang Network, one of China's earliest programmers to introduce the concept of "shareware"

This is a fundamental problem with "open weight" models where you don't know the training data. The biases are real, and they come out in really subtle ways.

flag

It's a genuine Chinese news article written by a real person. I don't know why they choose to use these types of images, it truly makes it look like slop. I think it's because it's originally a WeChat article posted on their website.
For what it's worth the original article is in Chinese: https://www.qbitai.com/2026/09/483600.html

The article linked here seems to be a (bad) auto-translated version.

It's not uncommon for images to be incorporated in an article like this, as they typically get shared around WeChat and correspond better with the 'culture'.

It's a big world, and I personally don't know anything about the history of software in places other than the west. Could you expand on the particulars of software in China and why those claims are wrong?
I can't speak to computing history in China, but Read Only Memory has books about the Japanese [0] and British [1] video game companies. MIT has published books about computing in Taiwan, Bulgaria, and China [2]. The former two are even open access, so you can download them for free.

[0] https://readonlymemory.com/products/japansoft-an-oral-histor...

[1] https://readonlymemory.com/products/britsoft-an-oral-history

[2] https://direct.mit.edu/books/search-results?f_ContentType=Bo...

What does any of this have to do with Chen Danyan?
I enjoyed this article... random cat pictures (esp. thumbs up cat picture) kept me going.

I do think on device AI will be a BIG thing... and basically change how we interact with our computers entirely - you will just ask to get things done. Low latency, free to use as much as you want. I think this future is hard to avoid/this will be a chunk of the future.

Like Star Trek -- "tea. earl grey. hot.". Though there will still be tasks where you'd have to likely want or need to pick up a keyboard and mouse. But of course there will be plenty of casual users who dont want or need to do anything interesting and use it for simple tasks like ask for the weather, turn on lights, and brew their earl grey.
For sure, there is a big subset of tasks where ultra low latency >>> other concerns
I don't think personal computers will be a BIG thing ....

- IDK, some random mainframe company exec in the 60s ;)

A dark horse named Lux.
Still censored, no thanks.
arent western models censored too ?
> At the 18th anniversary reunion of Shanda Innovation Institute, he publicly declared "eight non-consensus views for the AI era," ...

The "eight non-consensus views" matches a pattern in public communication in China, and I wonder if it is a technique that's taught, a style, or something with a history? Is there a word for it? I'm hoping someone here might know.

For example, Xi Jinping's English Wikipedia page includes:

* Four Comprehensives, Four Confidences, Four Consciousnesses, Three red lines, Three Stricts and Three Honests, Two Centenaries, Two Establishments, Two Upholds, Eight Musts, Eight-point Regulation

Deng Xiaopeng:

* Four Modernizations, Six Conceptions, Four Cardinal Principles, Four Modernizations

Mao and the Cultural Revolution:

* Three-anti and Five-anti Campaigns, Five Principles of Peaceful Coexistence, Four Olds, Five Black Categories, Five Red Categories, Four Greats, Three Loyalties and Four Limitlessnesses

Theres some strange affiniti for numberlike homynyms. You see similar in korean cultures. You would need to just find how the lexographics and spundslike to understand. Definitely doesnt survive transliteration.
Anyone know how this company expects to make money from AI model(s) running locally on consumer-grade hardware?

Download a closed-source binary for <platform>, payment via appstore? Ads inserted in output? Freemium vs. pro tier offerings? Some other way?

This seems like boxed software of old, right? Cracked and distributed on warez sites in no-time.

With specialized agents it could be as simple as "Upgrade your current used generic model to the better one specialized better on your tasks for only $10" rather then today's "we ship everything in one model no matter what you really need so you need our cloud subscription" approach.