Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Who's afraid of Chinese models?

stratechery.com

Post reply on HN
Print view View original

621–630 of 965 posts

Re: Who's afraid of Chinese models?

#621

Post by Jcampuzano2 »

tristanj wrote: post #611
causal wrote: post #583

"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.

1) Model distillation is the process of transferring knowledge from a large model to a smaller one. It doesn't require logits. https://en.wikipedia.org/wiki/Knowledge_distillation 2) The word "attack" is standard security vocabulary. Per RFC 4949: attack 1. (I) An intentional act by which an entity attempts to evade security services and violate the security policy of a system. That is, an actual assault on system se…

Sure that can be called an attack, but then we must also concede these labs essentially massively attacked everyone else in existence to get the data, and continue attacking as we speak.

In a way you could see this as a case of Robin Hood. The US companies exfiltrated all the data on the planet just to hoard it for themselves now and accuse anyone who tries to get a piece of that back from them, and the Chinese labs are distilling it to offer it for cheap.

Obviously a bit more complicated than that but it still holds pretty well.

Re: Who's afraid of Chinese models?

#622

Post by rarisma »

tristanj wrote: post #611
causal wrote: post #583

"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.

1) Model distillation is the process of transferring knowledge from a large model to a smaller one. It doesn't require logits. https://en.wikipedia.org/wiki/Knowledge_distillation 2) The word "attack" is standard security vocabulary. Per RFC 4949: attack 1. (I) An intentional act by which an entity attempts to evade security services and violate the security policy of a system. That is, an actual assault on system se…

So what kind of attack are AI companies doing by scraping up copyrighted info to build these LLMs?

"You're trying to kidnap what I've rightfully stolen."

Re: Who's afraid of Chinese models?

#623

Post by ralferoo »

sarchertech wrote: post #605
ralferoo wrote: post #599

Earlier quoted context omitted.

"big number bad" isn't a great way of analysing the data, because China has a massively larger population. The US emits 49% more CO2 per capita than China. And even with the much larger rate of increase, 0.79% to US' 0.3%, it'll still require 81 years to catch up to the US per capita rate [0]. The US aren't the good guys here, not by any means. Compared to my country, for instance, the US' emissions are 3x per capita…

In absolute numbers China produces 1/3 of the total worlds CO2, and they are still growing the amount of CO2 they are producing. That is not the actions of a country who “believes in climate change”. China’s CO2 per capita is ahead of every large developed country except for the US, Australia, and Russia.

You seem to think China is the evil polluter here. Just sort by per capita, and you'll see it's not even in the top 10% most polluting per capita.

There are plenty of large (depending on your definition of large) countries above it. Not just US, Australia and Russia that you are excluding for some reason, but also Qatar, Kuwait, UAE, Oman, Saudi Arabia, Canada all have significantly developed economically important countries. And they all have GDP per capita much greater than China's. What's your rationale for excluding them? There's a whole host of smaller countries with higher GDP per capita than China between those I listed above and China on the CO2 per capita results.

We should agree that ALL countries should be working to reduce CO2 emissions. But it sounds like you have a strong US bias and given Trump not only denies that climate change is even a thing, has withdrawn from international agreements on reducing emissions and even tries to pressure other countries to burn oil instead of investing in windfarms, it seems a bit disingenuous to try to make out that China is the only country that needs to improve. Before commenting on the speck in someone else's eye, first remove the plank from your own, yadda yadda...

Re: Who's afraid of Chinese models?

#624

Post by nixon_why69 »

sarchertech wrote: post #608
Ekaros wrote: post #601

Earlier quoted context omitted.

USA emits 2165574,97709 times more than Faroe Islands. Sounds like they should be forced to limit that to some reasonable level say 2x of that of Faroe Islands. Don't think it would be too big ask right?

China produces 1/3 of the world’s CO2. They already produce more per capita than all other large developed countries except Australia, the US, and Russia. And their rate of growth is higher than all of those countries except Russia. There’s no way to force them to do anything, but this is not that action of a nation that “believes in climate change”.

That's while being the world's factory, having lots of green energy, electric cars and an ACUTE motivation to reduce oil imports. Coal is a problem but at least they're not canceling half-finished green energy products when administrations change.

Re: Who's afraid of Chinese models?

#626

Post by Zigurd »

sarchertech wrote: post #605
ralferoo wrote: post #599

Earlier quoted context omitted.

"big number bad" isn't a great way of analysing the data, because China has a massively larger population. The US emits 49% more CO2 per capita than China. And even with the much larger rate of increase, 0.79% to US' 0.3%, it'll still require 81 years to catch up to the US per capita rate [0]. The US aren't the good guys here, not by any means. Compared to my country, for instance, the US' emissions are 3x per capita…

In absolute numbers China produces 1/3 of the total worlds CO2, and they are still growing the amount of CO2 they are producing. That is not the actions of a country who “believes in climate change”. China’s CO2 per capita is ahead of every large developed country except for the US, Australia, and Russia.

Actually, if you Google the rankings, China is 27th per capita, behind, of all places, Iceland.

Re: Who's afraid of Chinese models?

#627

Post by rektomatic »

1over137 wrote: post #580
throwaw12 wrote: post #486

Lets do "who's afraid of US models" version: * Me, as an individual, because I might not be able to pay price hikes, because my revenue (salary) is much lower than what they want and I can't support my expenses via huge bank loans. * Again, me as a new entrant to the industry, LLMs are basically pay-to-play games, again related to price hikes, new entrants might not be able to afford paying those prices 24/7 - which…

Also me, as someone who lives in Greenland, Canada, Venezuela, Cuba, Iran, etc. China is not threatening to invade, USA is.

you live in all those places??

Re: Who's afraid of Chinese models?

#628

Post by tedd4u »

pydry wrote: post #530
inglor_cz wrote: post #495

Earlier quoted context omitted.

While true, you neglect the fact that US corporations have some autonomy from the government. I doubt that any of the current American AI corporations get regular dossiers of desirable responses to sensitive political topics from Washington and implement them. That would be a reason for a major lawsuit and a major scandal. In China, the image of the country and the preferred narrative is under much tighter control of…

In China the state rules the corporations and in the US the corporations rule the state. Neither is a system worth fighting for. Swiss style democracy maybe but not autocracy and not oligarchy. >OTOH in Europe, US human rights policy was probably relevant in saving some dissenters in the former Eastern Bloc from torture or execution. Yeah, a bit like how Russia saved Edward Snowden. The US will use human rights as a…

The corps ruled the US until Trump. Now it’s moving to an oligarchical / kleptocratic mode like Russia. Sure the corps are still involved, but not on top. Will they remain after the current administration? I think it will be hard to put the genie back in the bottle.

Re: Who's afraid of Chinese models?

#629

Post by aqme28 »

tristanj wrote: post #611
causal wrote: post #583

"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.

1) Model distillation is the process of transferring knowledge from a large model to a smaller one. It doesn't require logits. https://en.wikipedia.org/wiki/Knowledge_distillation 2) The word "attack" is standard security vocabulary. Per RFC 4949: attack 1. (I) An intentional act by which an entity attempts to evade security services and violate the security policy of a system. That is, an actual assault on system se…

I'm not even convinced that this fits your definition. A distillation "attack" doesn't evade the security system in the sense of hacking past a login. The only part of the "security system" that it bypasses is the Terms of Service. And that's only after said data was acquired legally and correctly and normally.

It's a post-facto attack, which doesn't sit right linguistically to me.

Re: Who's afraid of Chinese models?

#630

Post by desterothx »

sarchertech wrote: post #605
ralferoo wrote: post #599

Earlier quoted context omitted.

"big number bad" isn't a great way of analysing the data, because China has a massively larger population. The US emits 49% more CO2 per capita than China. And even with the much larger rate of increase, 0.79% to US' 0.3%, it'll still require 81 years to catch up to the US per capita rate [0]. The US aren't the good guys here, not by any means. Compared to my country, for instance, the US' emissions are 3x per capita…

In absolute numbers China produces 1/3 of the total worlds CO2, and they are still growing the amount of CO2 they are producing. That is not the actions of a country who “believes in climate change”. China’s CO2 per capita is ahead of every large developed country except for the US, Australia, and Russia.

Almost like the entire worlds manufacturing was pushed to China the last 40 years (and is now getting moved to cheaper countries with the rising Chinese middle class). Is there a similar explanation for why the US number is so high?
Post reply on HN