CNN
–
Deepseek turned the world of high tech into his head last month. And according to artificial intelligence experts, they only see the beginning of the impact of Chinese tech startups on the AI field.
Deepseek grabbed the headlines in late January with the R1 AI model. The company says it can roughly match the performance of the open AI O1 model. Tech inventory fell as Deepseek temporarily released unpaid ChatGpt and became the top app on Apple’s App Store.
This achievement has led us to question the high-tech giants’ position in the AI competition with China, and the billions of dollars behind those efforts. Vice President JD Vance did not name Deepseek or China in his remarks at the Artificial Intelligence Litigation Summit in Paris on Tuesday, but he certainly does have a priority for the US to lead the sector. I emphasized whether there is.
“The United States is an AI leader and our administration has a plan to keep it that way,” he said, but he said, “The United States wants ‘partners’ with other countries. ” he added.
But it’s not just about Deepseek’s efficiency and power. The company’s decision to publish a critical portion of the technology, Deepseek R1, inferring and “thinking” through answers to deliver quality results, will also advance the field, experts say.
AI has been used for a long time in technology products, but has reached the flashpoint over the past two years thanks to the rise of CHATGPT and the rise in other generation AI services that have found how people work, communication and information. did. It ousted Wall Street darlings from companies like chip maker Nvidia, overturning the trajectory of the Silicon Valley giant. Therefore, developments that help build more competent and efficient models should be closely monitored.
“This is definitely not hype,” said Oren Etzioni, former CEO of Allen Artificial Intelligence Institute. “But this is a very fast world.”
Technology leaders responded quickly to the rise of Deepseek. According to CNBC, Google Deepmind CEO Demis Hassabis called the hype around Deepseek “exaggeration,” but also called the model as “probably the best job that’s coming out of China.”
Microsoft CEO Satya Nadella said in January’s quarterly revenue call Deepseek has “real innovations,” while Apple CEO Tim Cook said in an iPhone maker’s revenue call that “innovations that drive efficiency are the same.” It’s a good thing.”
However, all attention was not positive. Semianism, a semiconductor researcher, has questioned Deepseyk’s claim that training costs only $5.6 million. Openai told the Financial Times that it found evidence that Deepseek uses models from US companies to train its own competitors.
“We are aware and reviewing the indication that Deepseek may have inappropriately distilled the model and may share information as if we know the details,” Openai spokesperson told CNN It states in the comments. Deepseek could not be immediately contacted for comment.
And, as reported by Associated Press and ABC News, after security researchers highlighted potential links with the Chinese government, our pair of lawmakers already sought to ban apps from government devices. It’s there. Similar concerns have been raised about the popular social media app Tiktok, with the risk of being sold to American owners or being banned in the US.
“Deepseek is a (large language model) tiktok,” Etioni said.
The tech giant is already thinking about how Deepseek’s technology will affect products and services.
“What Deepseek gave us was essentially a recipe in the form of a technical report, but they didn’t give us any extra missing parts,” he said, providing tools for developers. said Lewis Tunstall, senior research scientist at Hugging Face, an AI platform. .
Tunstall is leading the effort to embrace Face in the fully open source DeepSeek R1 model. DeepSeek provided research paper and model parameters, but did not reveal code or training data.
Nadella said in Microsoft’s revenue call, Windows Copilot+ PCs, or PCs built to specific specifications to support AI models, can run locally distilled AI models from DeepSeek R1 . Mobile chip maker Qualcomm said Tuesday that models distilled from the Deepseek R1 are running on chip-equipped smartphones and PCs within a week.
AI researchers, scholars and developers are still investigating the meaning of Deepseek for AI advancements.
The Deepseek model is not only open source, nor is it the first model that could infer an answer before responding. Last year’s Openai’s O1 model can also do that.
What makes Deepseek important is how the AI community can infer and learn from other models, along with the fact that it is possible to see what’s going on behind the scenes. Those using the R1 model in the Deepseek app can see the “thinking” process when answering questions.
“You can see the wheels spinning inside the machine,” Durga Malladi, Senior Vice President and General Manager of Technical Planning and Edge Solutions at Qualcomm, told CNN.
Tunstall believes that we can see a wave of new models that allow inference like Deepseek in the not too distant future. As Tech Giants compete to build AI agents, Silicon Valley generally considers it to be the next evolution of chatbots and how consumers interact with devices, but that shift is still at all It’s not happening.
Grok 3, the next iteration of the social media platform X’s chatbot, has “very powerful reasoning capabilities,” and its owner Elon Musk, in a video appearance during the World Government Summit on Thursday I’ve said that.
For now, the AI community will continue to tinker with what DeepSeek has to offer. That is until the next breakthrough comes.
“I certainly expect to replace something else in the next 12 months,” Etioni said. “But that’s a very realistic progress.”
