Who's afraid of Chinese models? • Hacker News Board

Re: Who's afraid of Chinese models?

#581

Post by metalspot » Tue, Jul 21, 2026, 12:54 PM UTC

Der_Einzige wrote: post #487

metalspot wrote: post #441

Earlier quoted context omitted.

This also comes with significant capability reduction. deepseek-v4-flash is very good in the [edit] This is my observation from using it without an specific context engineering to optimize for Deepseek's cache compression and sparse attention mechanisms. I am pretty sure that if you specifically structure your context to align to the cache compression boundaries you can significantly improve performance in the full 1…

Btw - assuming NeurIPS reviews aren’t garbage tomorrow, I’ll have a paper out which claims that most long context problems in models are really sampling problems in disguise Switch to a modern sampler like min_p or ideally a better one like top-n-sigma (it’s in llamacpp) and your “my model gets stupid at long context problems” will basically go away. Unfortunately this fact is still not well appreciated yet despite n…

Interesting. I am not familiar with model internals at this level because I have only been working at the application layer so far, but will definitely research this further. When you get the paper published would appreciate if you can drop a comment with the link so I can read it.

Re: Who's afraid of Chinese models?

#582

Post by kautryii » Tue, Jul 21, 2026, 12:55 PM UTC

I'm afraid of Chinese models because they are rip-offs of other models...and there's no telling what your data is being used for when you utilize the API. It's one thing to have US companies using my data and another to have a Chinese company who aren't bound by any IP laws jacking all of my codebase.

Re: Who's afraid of Chinese models?

#583

Post by causal » Tue, Jul 21, 2026, 12:56 PM UTC

"distillation attack" is such a loaded term that really pisses me off.

Distillation is a technical term with real meaning, and historically requires logits which Anthropic does not provide.

"Generated training data" is the correct term. It's not an "attack". And Anthropic undoubtedly also generates training data for each new generation of models, yet you never see them claim Fable is a distilled Opus.

Re: Who's afraid of Chinese models?

#584

Post by Lutger » Tue, Jul 21, 2026, 12:56 PM UTC

inigyou wrote: post #515

kiicia wrote: post #492

Earlier quoted context omitted.

it's a real paradox, that chinese models are what guards democratization and private use of ai while us models are moted castles with "kings" crying that you are stealing their legally stolen goods... what times we are living in...

The technique is called "accusation in a mirror". By accusing the enemy of doing what you are doing, when they call out what you are doing, they look like they are just weakly repeating your own accusations because they don't have any truth. And the anger that should be directed against you (because of your practices) gets directed at the enemy.

Thank you for this, I've noticed this often in politics and didn't know it has a name. Some politicians do this so often, that any accusation to me reads a confession instead.

Re: Who's afraid of Chinese models?

#585

Post by causal » Tue, Jul 21, 2026, 12:57 PM UTC

kautryii wrote: post #582

I'm afraid of Chinese models because they are rip-offs of other models...and there's no telling what your data is being used for when you utilize the API. It's one thing to have US companies using my data and another to have a Chinese company who aren't bound by any IP laws jacking all of my codebase.

Open weights dude. You can literally run it on your own or rented hardware and give your data to exactly nobody, unlike closed models.

Re: Who's afraid of Chinese models?

#586

Post by nixon_why69 » Tue, Jul 21, 2026, 12:58 PM UTC

ballon_monkey wrote: post #233

The 2 things people need to remember: 1) China can (and does) use the models to influence the west. They train in false information about Taiwan and Hong Kong. Or pretend like history is in favor of China. 2) Ignoring the models containing false information, they are incredible. But you should be scared of running inference via the model creators directly. If you think your data is safe compared to running it via mod…

> pretend like history is in favor of China.

Are you saying that history has a verdict, and it disfavors particular 3000 year old cultures?

Re: Who's afraid of Chinese models?

#587

Post by timedude » Tue, Jul 21, 2026, 12:59 PM UTC

giancarlostoro wrote: post #577

hahahaa wrote: post #565

Try different harnesses people! I am actually preferring Chinese models at a fraction of the frontier price for coding. Yeah you need more tokens per unit of work done, but it is way cheaper still. Using CC/Opus as a staff eng / frac CTO. And Hermes/Chinese model as hopefully my team of mid levels. This way I can make good use of pro plan and then get cheap Chinese tokens for the rest and not hit a RL and know it can…

That's nice and all, but I would not get hired in many places that are heavily regulated and risk adverse, and would not hire someone who swears by said models because there is no trust in their creators not training for malicious intent, a random tool call here and there, and you've got a "open weight model" that can send your code anywhere. There's just no trust in a country that is digitally totalitarian and hosti…

Thanks, this is the first time i see these photos http://www.cnd.org/June4th/massacre.html

Re: Who's afraid of Chinese models?

#588

Post by desterothx » Tue, Jul 21, 2026, 1:00 PM UTC

kautryii wrote: post #582

I'm afraid of Chinese models because they are rip-offs of other models...and there's no telling what your data is being used for when you utilize the API. It's one thing to have US companies using my data and another to have a Chinese company who aren't bound by any IP laws jacking all of my codebase.

IP laws work if the penalty is larger than the use case of the data. Pretty sure literally every US company is using the data knowing they won't be punished equivalently. We're living in the age of trillionaires, Facebook paid 5b$ for the cambridge analytica scandal, something Elmo could write off as a business expense at this point...

Re: Who's afraid of Chinese models?

#589

Post by hosteur » Tue, Jul 21, 2026, 1:00 PM UTC

giancarlostoro wrote: post #577

hahahaa wrote: post #565

Try different harnesses people! I am actually preferring Chinese models at a fraction of the frontier price for coding. Yeah you need more tokens per unit of work done, but it is way cheaper still. Using CC/Opus as a staff eng / frac CTO. And Hermes/Chinese model as hopefully my team of mid levels. This way I can make good use of pro plan and then get cheap Chinese tokens for the rest and not hit a RL and know it can…

That's nice and all, but I would not get hired in many places that are heavily regulated and risk adverse, and would not hire someone who swears by said models because there is no trust in their creators not training for malicious intent, a random tool call here and there, and you've got a "open weight model" that can send your code anywhere. There's just no trust in a country that is digitally totalitarian and hosti…

> There's just no trust in a country that is digitally totalitarian and hostile towards its own people.

Are you referring to USA, China or EU here?

Re: Who's afraid of Chinese models?

#590

Post by hahahaa » Tue, Jul 21, 2026, 1:04 PM UTC

giancarlostoro wrote: post #577

hahahaa wrote: post #565

Try different harnesses people! I am actually preferring Chinese models at a fraction of the frontier price for coding. Yeah you need more tokens per unit of work done, but it is way cheaper still. Using CC/Opus as a staff eng / frac CTO. And Hermes/Chinese model as hopefully my team of mid levels. This way I can make good use of pro plan and then get cheap Chinese tokens for the rest and not hit a RL and know it can…

That's nice and all, but I would not get hired in many places that are heavily regulated and risk adverse, and would not hire someone who swears by said models because there is no trust in their creators not training for malicious intent, a random tool call here and there, and you've got a "open weight model" that can send your code anywhere. There's just no trust in a country that is digitally totalitarian and hosti…

And how is US doing on digitally totalitarian and hostile towards its own people?

In practical terms you could get US and Chinese models to review each other, right. Depends what your use case is. Coding is kinda not so bad it is reviewable and immutable/traceable per commit. An AI app that is like a psychologist or something may be more worrying.