Technical leaders reply to the speed of deepsek

If you haven't heard, a new AI stands in the city: A deep viewThe subscription of Volunitary Analysis (quantity) is based in Hong Kong, high-flyer Capital Management, on Promoted waves over Silion Valley and the great world with a Discharge earlier a week of new open taxi modifance, Deepsek R1which is equal The most powerful openai model – And at a fraction of your consumer cost and company (when it trained).

While DEPSEK R1 already reschedreedted on a regularly Turving Marksy Turvy, shifting fast and quite competitively with Anthropic and Google for the most powerful property modules, and Met Platforms came in this time that the company is behind a 'A hot model based in China, the geology “Frenemy“US, and their technical region was widely regarded, to this period, lower than those at Silion Valley.

Therefore, it has not caused a shortage of handwriting and anniversary from the US and West 'block technologies, which are suddenly doubt for more money. throw and more computing (Graphics), GPUS, the most commonly powerful game of casual). used to train ai modules) toward the problem of creating becoming more powerful models.

But some spectacular leaders on a public response is to a very hopeful climbing of Deepsek.

Marc Andreesen, an engineer of the newest of mosaic web browser, netscape browser company and Current General Partner at the Andreesen horowitz partner (A16z) Enterprise Capital, posted on x today: “Deepsek R1 is one of the most amazing and most impressive supplicity of the most spectacular and exposure to the world ever seen – and as a source of open, a deep gift to the world (robji emoji).”

Yann Lucun, the senior scientist AI for the Meta's casted AI Research Department (Fair), up on a Linkedin account:

“To people who see a deepsek performance and think:
'China is over the USA in Ai.'
You're reading this wrong.
The correct reading is:
'A code with an open code crossing owner-occupier models.'

Deepzek has benefited from open search and open store (eg pytorch and llama from meta)
They came up with new ideas and took them as well as the work of others.
As their work has published and open store, everyone can benefit from.
That is an open search and open source power. “

And even Mark “Zuck” Zuck “Zuck” Zuck's Meta AI, was probably founder and head of Meta Ai, trying to issue a Deepzek raising with his one. post itself on Facebook promising that a new version of ai of ai of Facebook AI's Facebook Mail Facebook as “Proper Model” when released once this year. As sent ee:

“This will be a special year for AI. In 2025, I expect Meta Ai as the lead to be served over 1 billion people, Llama 4 will be the main modern process which begins ai engineer that starts ai engineering to start adding more Code to our R & D. efforts. To power this, Meta builds a 2GW + data center that covers an important part of manhattan. We will: 1gw of online computer in '25 and complete the year with more than 1.3 million GPU. We expect to investing $ 60-65b in Capex this year and we also grow significantly our team teams, and our capital to continue to invest in the coming years. This is a great effort, and over the coming years it will be leading our key factors and industry, resolving the American innovation, and extending the American innovation. Let's go up!“

He even shared the Dataacenter 2 Gigawatt which was mentioned in Manhattan:

Clearly, even as it is involved in enthusiasm with an open code, Zuck is not sure to make the best of efficiency for efficiency Gpus than the main work of the right one for meta, or for the future of Ai.

But with US companies increasing and / or Tits high amounts of amounts of a new AI infrastructure that many experts have declinedly decrease (due to hardware / software advances, yet which the bird is still winning future to be won the main provider AI for the world. Or maybe you may be diversity of models every man with a smaller market section? Keep eye, because this contest becomes closer and stronger than ever.