Omer Taha Cetin | Anadolu | Getty Images
Deepseek’s powerful new artificial intelligence model is not just a win for cinema-this is a victory for an open technology code version than similar TargetDatabricks, Mistral and hug, according to industrial experts who spoke with CNBC.
Last month Deepseek posted R1, An open code reasoning model that claims that Openi’s O1 models rival Using a cheaper, less energy intense process.
Development caused the market values of NVIDIA and other chips manufacturers falling for fear that it could lead to reduced spending on high-performance computer infrastructure.
Deepseek is a Chinese AI laboratory that focuses on the development of large language models with the ultimate goal of achieving artificial general intelligence or agi. It was founded in 2023 Liang Weenfeng, co -founder of a quantitative hedge fund aimed at AI.
AGI is slightly related to the idea of Ai that is equal to or outweighs the human intellect on the wide range of tasks.
What is the open code ai?
Since Openai -ov Chatgpt broke out on stage in November 2022, researchers AI working hard to understand and improve the progress of the basic technology of a large linguistic model that supports it.
One area of focus for many laboratorys was an open code AI. Open source refers to software whose original code is freely available on the open web for possible changes and redistribution.
Many companies from technological giants such as Meta to Scrappier startup, such as Mistral and face hugs, bet open source as a way to improve technology, while sharing important events with wider research communities.
As Deepseek empowered an open source
Deepseek’s technological breakthrough has only done the case for AI models of open code stronger, according to some technical executives.
Seeda Rejal, Chief Commercial Director AI startup Netmind, told CNBC that the success of the Chinese company shows that the AI open code AI “is no longer just a commercial research initiative, but a sustainable, scalable alternative to closed models” like Openi’s GPT.
“Deepseek R1 has shown that open code models can achieve top performance, rival ownership models from OPENAI and others,” Rejal told CNBC. “This provokes the belief that only closed source models can dominate innovation in this area.”
Rejal is not alone. Yann Lecun, the main AI scientist, said Deepseek’s success represents a victory for AI models of open code, not necessarily a victory for China over the United States. The target is behind the popular AI model of an open code called Llam.
“People who see the performance of Deepseek and think,” China surpasses now in Ai. “You’re reading this wrong.
“Deepseek profited from open research and open code (eg Pytorch and Llam from Meta). They came to new ideas and built them on top of other people. Since their work was published and an open source, everyone can profit from it. It is the power of open research and open code. “
Open Koker AI goes globally
Cut off Washington from access to advanced chips needed to train and start AI models, cinema turned to open code technology to increase the attraction of its AI models. Many Chinese companies – included Deepseek – implement open code models as a way to increase innovation and spread their use.
But a trend of companies that addresses the technologies of open code for success in AI is not limited to China. In Europe, the Association of Academics, Companies and Data Centers united in the development of families of high quality, multilingual large language models called OpenUurollm.
The Alliance runs Jan Hajic, a well -known computational linguist at Charles University, Cehija, and Peter Sarlin, co -founder of Silo AI, AI laboratory who bought the US AMD chips last year.
The initiative is part of the wider pressure on “AI sovereignty”, in which countries encourage investing in their own domestic AI laboratory and data centers to reduce the reliance on the Silicon Valley.
What is the catch?
However, there are disadvantages of open code AI. Experts warn that, although open -code technology, it is a good thing for innovation, it is also more prone to cyber exploitation. This is because anyone can prepare and modify it.
Cibernetic security companies have already discovered vulnerability in Deepseek’s AI models. A study that Cisco announced last week revealed that R1 contained critical security defects.
Using the “Jail Algorithmic Technique, the CISC’s AI security research team says he received R1 to provide affirmative answers to a number of harmful instructions of the popular Harmbench” with the rate of 100 % success of the attack. “
“Deepseek R1 was allegedly dressed with a fraction of budgets that other border models provides on the development of their models. However, this comes at a different expense: safety and security,” Cisca researchers Paul Kassian and Amin Karbasi wrote.
Data missing is also concerned. Data processed by Deepseek R1 models via his website or application are sent straight to the cinema. Chinese technological companies have long been swallowed by the claims that Beijing is used by the spying systems of western entities and individuals.
“Deepseek, like other generative AI platforms, is a double-blade sword for companies and individuals,” said Matt Cooke, a strategist of Cyber-security EMEA on Proolpoint. “Although the potential for innovation is indisputable, the risk of data leaks is serious concern.”
“Deepseek is relatively new and will take time to learn about technology; however, what we know is to enter the sensitive data of the company or personal data into these systems such as the surrender of a loaded weapon attacker,” Cooke added.
Netmindov Rejal told CNBC to AI Open Code Models Introducing the risks for cyber security that companies should take into account, including attacks of chain of software supply, rapid events in eggs and so -called “data poisoning” that try to introduce bias or harmful results.
WATCH: Why Chinese Deepseek brings American Ai leadership in danger
