Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
OMER TAHA CETTIN | Anadolu | Getty images
The powerful artificial intelligence model of Deepseek is not just a victory for China: it is a victory for open source versions of the technology of the tastes of GoalDatabricks, Mistral and hugging the face, according to industry experts that spoke with CNBC.
Last month, Depseek launched R1, a Open source reasoning model that claims to rival OPENAI O1 model performance using a cheaper and less energy intensive process.
Development caused NVIDIA market values and other chip manufacturers to collapse due to the fear that it could lead to a reduced expense in high performance computer infrastructure.
Deepseek is a Chinese laboratory that focuses on developing large language models with the ultimate goal of achieving artificial general intelligence, or AGI. It was founded in 2023 by Liang Wenfeng, co-founder of the quantitative coverage fund focused on High-Flyer Fund.
AGI freely refers to the idea of an AI that matches or overcomes the human intellect in a wide range of tasks.
What is open source AI?
Since the OpenAI chatgpt broke into the scene in November 2022, IA researchers have been working hard to understand and improve the advances in the fundamental large language model technology that supports it.
An approach area for many laboratories has been open source AI. Open Source refers to software whose source code is available free on the open web for possible modification and redistribution.
Many companies, of technological giants, as a goal to Starpier Startups, such as Mistral and Hugging Face, have been betting on open source as a way to improve technology while also sharing important developments with the research community in general.
How Deepseek empowered open source
The Depseek technological advance has only made the case of the strongest open source AI models, according to some technology executives.
Severa Rejal, commercial director of the startup of IA, Netmind, told CNBC that the success of the Chinese firm shows that open source AI “is no longer only a non -commercial research initiative, but a viable and scalable alternative to the CLOSED MODELS “LIKE THE OPENAI GPT.
“Deepseek R1 has shown that open source models can achieve state -of -the -art performance, rival OpenAi patented models and others,” Rejal told CNBC. “This challenges the belief that only closed code models can master innovation in this space.”
Rejal is not alone. Yann Lecun, the head scientist of Meta, said that Deepseek’s success represents a victory for open source AI models, not necessarily a victory for China over the United States. Goal is behind a popular open source AI model called flame.
“For people who see Deepseek’s performance and think:” China is surpassing the United States in AI. “You are reading this badly.
“Deepseek has benefited from open research and open source (for example, Pytorch and Meta flame). They came up with new ideas and built them on other people’s work. Because their work is published and open source , everyone can benefit from him.
Global open source ai
Washington cut to access advanced chips necessary to train and execute AI models, China has resorted to open source technology to increase the appeal of its AI models. Many Chinese companies, including deep bases, are looking for open source models as a way to increase innovation and spread their use.
But the trend of companies that resort to open source technologies for success in AI is not limited to China. In Europe, an alliance of academics, companies and data centers has been associated in the development of a family of large high performance multilingual language models, called Openeurollm.
The alliance is directed by Jan Hajič, a recognized computational linguist at the University of Charles, Chechia, and Peter Sarlin, co -founder of Silo AI, an AI laboratory that was bought by the American chip manufacturer AMD last year.
The initiative is part of a broader impulse for the “sovereignty of AI”, in which countries are encouraging investment in their own laboratories and national data centers of AI to reduce a dependency of Silicon Valley.
What is the capture?
However, there are disadvantages for open source AI. Experts warn that, although open source technology is a good thing for innovation, it is also more prone to cyber exploitation. This is because anyone can replace and modify it.
Cybersecurity companies have already discovered vulnerabilities in Depseek’s models. The investigation published last week revealed that R1 It contained critical security failures.
Using “Jailbreak algorithmic techniques”, the Cisco’s IA security research team says that R1 would provide affirmative responses to a series of harmful indications of the popular Harmbench “with a 100%attack success rate.”
“Deepseek R1 was supposedly trained with a fraction of budgets that other border models suppliers spend on developing their models. However, it has a different cost: security,” wrote Cisco Paul Kassianik and Amin Karbasi researchers.
Data escape is also a concern. Data processed by Deepseek’s R1 model through its website or application are sent directly to China. Chinese technological firms have been persecuted by accusations that Beijing uses their systems to spy western entities and individuals.
“Deepseek, like other generative AI platforms, presents a double -edged edge sword for companies and individuals equally,” said Matt Cooke, cybersecurity strategist Emea in Proofpoint. “While the innovation potential is undeniable, the risk of data leakage is a serious concern.”
“Deepseek is relatively new, and it will take time to learn about technology; however, what we do know is to feed confidential data of the company or personal information in these systems is how to deliver to the attackers a loaded weapon,” added Cooke.
Netmind Rejal told CNBC that open source AI models introduce cyber security risks that companies should consider, including attacks in the software supply chain, fast progress and so -called “data poisoning” events events ” They try to introduce harmful biases or results.
LOOK: Why China’s Speek is to endanger the leader of the United States.
