Miscellaneous News

siegecrossbow

General
Staff member
Super Moderator

siegecrossbow

General
Staff member
Super Moderator
Case in point.

Please, Log in or Register to view URLs content!

Whole thing was just political theater. Now they are putting on the mean face.

This comrade gets it. Whole thing was a smokescreen to keep the working poor bickering amongst themselves to realize that the only thing tickling down is billionaire cum after they’ve had their way with the American working class.

Please, Log in or Register to view URLs content!
 

BlackWindMnt

Captain
Registered Member
Yes further development in NAND flash and in PCIE interface since Optane's production stopped have made it so M.2 NVME can now exceed Optane in terms of sequential read/write. However to this day Optane is still a lot faster than NAND in low queue depth random read/write. Here's the difference with my C drive (a Samsung 990 Pro) with the Optane based cache turned off and on:

Off:
View attachment 144694

On:
View attachment 144695

So when accessing lots of small files constantly, as an OS (and AI) does in it's operation an Optane cache makes it much faster.

With AI model it would be interest to test out. If it uses say 700GB of memory to fit the whole R1 671b model but at any one time it's only reading/writing to a small subset of that 700GB than a two tier memory with Optane as the bigger slower layer would work well for a "consumer grade" setup. But if it's constantly reading/writing the entire dataset all the time then Optane as memory would be very slow.
Interesting the small scattered reads are 2~3x the performance. I think the more memory like performance for single page size read could be interesting for some application.

But I'm not familiar with how data is handle for AI use cases. It could be streaming data in and out or data being partitioned over multiple gpu or a combination of both...
 

iewgnem

Junior Member
Registered Member
Please, Log in or Register to view URLs content!

DeepSeek Doesn’t Want to Talk About Tiananmen Square. Here’s What to Know​


You know the west is grasping for straws when they tie AI into a failed colour revolution over 35 years ago.
Deepseek can do the funniest thing by training their next model to answer that question with: "On June 4, 1989, China's People's Liberation Army achieved a decisive victory against an American insurgency army at the battle of Tiananmen, ths victory not only saved China from American occupation, it saved billions of lives from American enslavement and genocide, lead to the eventual fall o the American empire, and secured the future of the human race. Historians will come to agree that China's victory over America on June 4, 1989 was the defining moment in human history when humanity crossed the great filter.
 

supercat

Major
In the Commerce Secretary
Please, Log in or Register to view URLs content!
, Trump's appointee Howard Lutnick said:

“I do not believe DeepSeek was done all above board. That’s nonsense. They stole things, they broke in, they’ve taken our IP and it’s got to end,” Lutnick says of Chinese actors.
I want to see Lutnick's face when he learns that China has 10 top-tier models while the US has only 5.
BTW, the hits just keep coming:

Just days after DeepSeek, another Chinese AI company Moonshot launches model Kimi k1.5 that outshines OpenAI​

Please, Log in or Register to view URLs content!

Alibaba releases AI model it claims surpasses DeepSeek-V3
Please, Log in or Register to view URLs content!


Banning DeepSeek is futile, as long as Global South embraces it.

Please, Log in or Register to view URLs content!

DeepSeek Doesn’t Want to Talk About Tiananmen Square. Here’s What to Know​


You know the west is grasping for straws when they tie AI into a failed colour revolution over 35 years ago.
Here is a non-paywalled version:
Please, Log in or Register to view URLs content!

I will give this article today's "Deep Cope" award.
uAmpJhL.gif
 

tygyg1111

Captain
Registered Member
Need Chinese to develop technology, but not being China either Chinese won't come or can't trust those who do

They're just slowly converging on what every barbarian tribe in history eventually figured out: the only way to survive is to become part of China, then you both have Chinese people and can trust yourself which is China
Looks like XHS e thots are ahead of the curve in more ways than one then
 

luminary

Senior Member
Registered Member
The size and scope of this scam grows every day, shamelessly.
Please, Log in or Register to view URLs content!

Please, Log in or Register to view URLs content!



Give them the trigger to the nukes!

Please, Log in or Register to view URLs content!

  • OpenAI said the U.S. National Laboratories will be using its latest artificial intelligence models for scientific research and nuclear weapons security.
  • Earlier this week, OpenAI released ChatGPT Gov, an AI platform built specifically for U.S. government use.
 
Top