Artificial Intelligence thread

T

tokenanalyst

Lieutenant General
Registered Member
tamsen_ikard said:
You overestimate chinese govt too much.
Click to expand...
I don't.
tamsen_ikard said:
No one except the chinese AI startups have the talent and knowledge to train a model with trillion parameters, especially this fast. AI is changing too fast for lumbering state institutions to keep up.
Click to expand...
Do you think the research is done by these companies is in vacuum or a lot of the research is done in collaboration with China research institutions? The research literature shows a lot is done in collaboration with China universities. I know that Moonshot, Ziphu and Deepseek where purely academic projects before becoming commercial projects.

Also China institutions have their own datacenters or have access to datacenters. Is very likely they can take the DeepSeek, or Ziphu or Moonshot model architecture along with the secret sauce and train their own multi trillion parameters models for cyber or military data processing applications. That is not counting the finetuning of these models for specific applications.

tamsen_ikard said:
And those AI startups dont have the bandwidth to keep up with US SOTA and develop completely hidden trillion parameter models.

Even US govt do not have hidden trillion parameter model. They r also completely reliant on whatever US startups can develop.
Click to expand...
Is fair to said that China has more than enough computing power to train models.
 
Last edited:
T

tokenanalyst

Lieutenant General
Registered Member
bsdnf said:
Government departments and state-owned enterprises are most actively deploying LLM locally for paperwork processing.

In my view, they have neither the capability nor the willingness to do this. LLM has advanced too rapidly; they prefer to reap the benefits without contributing anything
Click to expand...
Some government agencies will tokemaxx from DeepSeek, GLM or Kimi Cloud API services or these companies will implement local AI servers.
But giving how China security agencies and military operates, they absorb technology and the implement their own specific made versions, is already happening in robotics and drone technology. Is not out of the realm of possibility.
 
H

henrik

Captain
Registered Member
Please, Log in or Register to view URLs content!
tamsen_ikard said:
You overestimate chinese govt too much. No one except the chinese AI startups have the talent and knowledge to train a model with trillion parameters, especially this fast. AI is changing too fast for lumbering state institutions to keep up.

And those AI startups dont have the bandwidth to keep up with US SOTA and develop completely hidden trillion parameter models.

Even US govt do not have hidden trillion parameter model. They r also completely reliant on whatever US startups can develop.
Click to expand...

You underestimate the Chinese government too much. They have hidden LLMs totally separated from outside the government. These models are designed by the top talents skimmed off by the government straight from the top schools.
 
bsdnf

bsdnf

Senior Member
Registered Member
In an interview, the head of MINIMAX research stated that their fable-calss model, the M3, will be released within six months.

Kimi K3 has already been released, and Deepseek V4 Pro, also a fable-class, is about to be officially released. Z.ai's GLM-5.5 is rumored to be released next month as well. "Within 6 months" means they are way behind schedule.
 
You must log in or register to reply here.
Top