This source claims inference in around 38 t/s, which is not bad:
Artificial Intelligence thread
- Thread starter 9dashline
- Start date
This source claims inference in around 38 t/s, which is not bad:
"Scientists from the two countries have developed a breakthrough algorithm using information from reverse engineering video card accelerators. The algorithm allows gaming GPUs to be used for scientific computing[...].
The innovation was achieved by specialists from Shenzhen MSU-BIT University, co-founded by Lomonosov Moscow State University and Beijing Institute of Technology[...]It also means that Russia and China need to buy less NVIDIA GPUs[...]
These advancements highlight Russia’s technological expertise, allowing collaboration with China and potential future partnerships with India." Sputnikintl
The innovation was achieved by specialists from Shenzhen MSU-BIT University, co-founded by Lomonosov Moscow State University and Beijing Institute of Technology[...]It also means that Russia and China need to buy less NVIDIA GPUs[...]
These advancements highlight Russia’s technological expertise, allowing collaboration with China and potential future partnerships with India." Sputnikintl
Not sure what to make of this tweet from DeepSeek AI researcher Guo Daya:
manqiangrexue
Brigadier
They want Chinese people to benefit the US and NOT benefit China; that's their stupid logic. They used to welcome China's brightest minds. The result is that those minds came, learned,benefitted America but a large portion went back and did much more for China. So American politicians closed America to Chinese minds and the result is that Chinese scientists just worked for China and the US got left behind. Now they want to go back to inviting Chinese minds cus they already forgot why they closed off in the first place and that started just 2 years ago. That's how smart these people are... It's like a guy going back and forth cus he can't decide if he prefers to get stabbed or shot to death and he has really bad memory so it's kinda funny cus he looks surprised every time.but but but USA has advantage in ''attracting global talent''
US should ‘steal’ China’s best AI talent to keep pace, Senate hears
The US should welcome China’s best scientific minds into its universities to compete with the mainland’s success in , American lawmakers in Washington heard on Thursday, as Chinese start-up this week.
“Let’s steal their best engineers,” said Melanie Hart of the Washington-based Atlantic Council at a hearing convened by the Senate Foreign Relations Committee.
View attachment 144745
Means their next R2 will be a lot smarter, probably beating o3
Okay, this is cheaper than average corporate laptop now.....
They should wait for OpenAI to release their full o3 model before, and then come out with R2 for maximum impact.
Regarding Chinese students/researchers in US, this is another 3D chess genius move: it is a very subtle way to workaround embargoes.
Once in US these people can use plenty of hardware for their research, including quantitative tests of different algos, architectures, etc...all stuff that requires huge amount of hardware resources. Then they publish their results for everybody to see and learn and follow up directly without very expensive trial and error.
All Chinese big players, DeepSeek included, have greatly benefited and will benefit from research papers published in US by Chinese authors. So on one hand they give prestige (and indirectly money) to their US Universities/institutes, OTH they help to push the SOTA for all the world to benefit.
It is like: You don't want me to use your GPUs? I come to your home and use them from there. I let you put your institute name on my paper, but as long as you allow me to publish, this is ok for me...and good for the world.
It would be different if the research carried on by Chinese scientists in US would not be allowed to be published, but in those cases, like for instance for defense/military stuff, Chinese scientists are kept out anyway, due to security clearance rules. So all that stuff is made by US citizens without support from Chinese researchers.
Sam Altman: OpenAI has been on the ‘wrong side of history’ concerning open source
"a different open source strategy"
I think the OpenAI brand is forever dead, they should just close the org and restart a new org with the same people but a different name.
Last edited:
DeepSeek is recruiting for positions in Beijing. Salaries for "core system R&D" engineers freshly out of college range from 60k to 90k yuan, roughly 9k to 13k USD.
据红星资本局报道，深度求索公司的北京子公司目前正在招聘 52 个岗位，涵盖深度学习研究员、核心系统研发工程师及资深 UI 设计师等职位，均为 14 薪。其中，核心系统研发工程师（校招）的薪资范围为 6 万元至 9 万元，大部分岗位的起薪在 2 万元以上。公司还提供高薪实习机会，AGI 大模型实习生的日薪为 500 元至 1000 元，数据百晓生实习生为 500 元 / 天-510 元 / 天。
据红星资本局报道，深度求索公司的北京子公司目前正在招聘 52 个岗位，涵盖深度学习研究员、核心系统研发工程师及资深 UI 设计师等职位，均为 14 薪。其中，核心系统研发工程师（校招）的薪资范围为 6 万元至 9 万元，大部分岗位的起薪在 2 万元以上。公司还提供高薪实习机会，AGI 大模型实习生的日薪为 500 元至 1000 元，数据百晓生实习生为 500 元 / 天-510 元 / 天。
