Join our daily and weekly rapporteurs and get the latest updates and exclusive content of top AI coverage. More information
If you haven’t heard, a new star of artificial intelligence is in the city: Deepseek, a subsidiary of Hong Kong’s quantitative analytical (quantitative) company High-Flyer Capital Management, sending shock waves all over Silicon Valley and world. This week of the new Open Source model of a large reasoning model, Deepseek R1, which corresponds to the most powerful available model Openai O1 – and for a fraction of the cost of users and for society itself (during its training).
Although Deepseek R1 has already transformed a permanent, fast -moving and intensively competitive market for new artificial intelligence models – in previous months, Openai jockey with Anthropic and Google on the most powerful proprietary models, while Meta Platforms often came up with “close” opponents With an open source code – this time the difference is that the company standing behind the hot model is located in China, the geopolitical “mad” of the US, and whose technology sector has been widely considered to be inferior. to Silicon Valley.
As such, the lack of manual squeezing and existentialism from the US and Western bloc technicians who suddenly doubt the OpenNei and the general large technological strategy to throw more money and more computer technology (graphic design units, GPUs, powerful gaming chips usually used to train models artificial intelligence) towards the problem of inventing increasingly powerful models.
Yet some Western Tech leaders had a predominantly positive public response to Deepseek’s rapid rise.
Marc Andreessen, a co -star of the Mosaic web browser, co -founder of the Netscape browser and the current general partner of the famous Risk Capital Society Andreessen Horowitz (A16Z), today wrote: Source deep gift to the world (robot emodi, emoticon greeting). ”
Yann Lecun, the main scientist AI for Meta’s Fundamental AI Research (Fair) Division, published on his LinkedIn account:
“For people who see Deepseek’s performance and think:
“China surpass the US in AI.”
You’re reading it wrong.
The correct reading is:
“Open source models outperform those proprietary.”
Deepseek benefited from open research and open source (eg Pytorch and Llama of Meta)
They came up with new ideas and built them on the work of other people.
Because their work is published and has an open source, everyone can benefit from it.
This is the power of open research and open source. ”
And even Mark “Zuck” Zuckerberg, founder and Executive director of META AI, seemed to be trying to face the ups of Deepseek with his own Facebook post, which promises that the new version of the AI family with the open source code will be “Art leading Model ”when it comes out sometime this year. As he said:
”This will be a decisive year for AI. I expect Meta AI to be the main assistant to serve more than 1 billion people in 2025, Llama 4 will become the leading model of the most modern technology and build engineers AI, who will contribute more and more code to our research and development. effort. To this end, the META 2GW+ data center, which is so large, builds a significant part of Manhattan. In 1925, we will open up online ~ 1GW computer technology and end the year with more than 1.3 million GPUs. This year we plan to invest $ 60-65 billion in capital investments and at the same time significantly expand our AI teams and we have capital to continue investment in the coming years. It is a massive effort and will manage our main products and business in the coming years, unlock historical innovations and expand American technological leadership. We’re going to build!”
He even shared graphics depicting 2 gigawatt data center mentioned in his post on Manhattan:
Obviously, although Zuck holds a commitment to open source AI, it is not convinced that the approach Deepseek optimization for efficiency in using much less GPUs than large laboratories is the right one for meta or the future of AI.
But since American companies receive and/or spend record amounts on a new artificial intelligence infrastructure that, according to many experts, is rapidly degraded (to improve hardware/chip and software), the question remains that vision will eventually win and become dominant Artificial intelligence provider. Or maybe there will always be more models, each with a smaller market share? Stay tuned because this competition is getting closer and fierce than ever.
