pull down to refresh
I have not seen that change reported but you may be correct.
China does have an advantage in its huge lead in electricity power supply and cost.
USA faces power shortages, price hikes and rolling blackouts if it is to power the projects data centres.
China’s AI Power Play: Cheap Electricity From World’s Biggest Grid- WSJ
https://archive.ph/OrJVm
We're following sources a bit closer to the fire here than WSJ that tries to comfort normies about some thing they don't understand...
#1421164 is the last time we talked about this.
It sounds like you didn't read the WSJ article- it is not comforting. It is however factual, informative and detailed analysis.
Your link is about as shallow and narrow an analysis of a complex economic and strategic issue as can be imagined.
I guess you were too busy C&P-ing shit and downzapping, and didn't care to have the conversation before. We could have had it...
We're not in ~Politics_And_Law here though, we're in ~AI. So we assess the products and the actual results, not so much what people are inferring that the CCP wants or does not want. Do you think the CCP wanted to have tens of millions of half-finished cars that sucked all the available subsidy? Come on... there's lots of things that happen in China that is not what "the CCP" wants.. hierarchically. Doesn't mean it doesn't happen.
Back on topic, both Ali and Moonshot have switched from 100% open weights to the Google model of proprietary high-end + open lesser models. Deepseek has had a couple unreleased high-end models too, and only Z.ai has been consistently releasing open weights. How do you explain that from the WSJ article? Does it even mention what the leading Chinese LLM is doing? Or what the leading Chinese model is in the first place? No, it doesn't. It only mentions Deepseek. Why? That's been bested by all 3 main Chinese competitors 8 months ago.
The article tries to distinguish some high level direction to figure out what state is winning in a war that is waged in symbolics only. On the ground all we see is massive progress despite that war. Z.ai (GLM) is doing awesome lately and exclusively training on Huawei Ascend (and thus benefitting from the energy abundance). Moonshot (Kimi) and Deepseek are training exclusively on Nvidia. Ali (Qwen) is unconfirmed to be hybrid. The landscape, also within China, is diverse. There is no single AI strategy. Not in China, not in the US. The strategy of the politicians on both sides of the pacific is to "win", except with AGI gone, no one can tell what they're racing towards.
Would much prefer to have constructive dialogues and contests of ideas than slanging matches.
But it takes two to tango.
You are right- the CCP cannot dictate everything as in a classic command economy but it does set and direct overall strategic directions and does control capital allocation to such strategic priorities.
AI is definitely such a priority where it has set a different strategy to the US/West.
I do not know as much about AI technical specifics as you clearly do but it is one of the fronts of this current contest between the US and China and must be seen in that context to be fully understood.
In some areas such as the massive quantities of electricity required to power the data centres required the CCP strategic direction is clearly a major factor and advantage for China.
Chinese and US AI platforms working together may well be happening- the supply of the best US chips to China seems to be a significant US bargaining chip.
The Chinese approach as I have seen it described is to utilise AI for projects that enhance Chinese economic efficiency now and in the near future in contrast to the US approach of seeking AI consciousness as some sort of holy grail that will preserve US hegemony and dominance.
You say 'except with AGI gone'.
Have the US AI companies given up on achieving AGI, or do you see it as unachievable?
The Chinese strategy as I understand it never depended upon AGI being achieved/achievable but rather focuses on what can be achieved short of that.
Wow ok was not aware of that.
I thought a lot of their talk/hype was based upon the premise of achieving AGI.
Need to do some more study of this.
That was until the promised almost-AGI (gpt-5) kinda was a doozy.
Those that are still working on "Super Intelligence" are now mostly outside of the usual suspects and not working on LLMs anymore, check for example LeCun (#1405916):
“I’m sure there’s a lot of people at Meta, including perhaps Alex, who would like me to not tell the world that LLMs basically are a dead end when it comes to superintelligence,” he says. “But I’m not gonna change my mind because some dude thinks I’m wrong. I’m not wrong. My integrity as a scientist cannot allow me to do this.”
It does both now