Photo credit: www.cnbc.com
DeepSeek’s recent advancement in artificial intelligence signals a notable shift in the tech landscape, benefiting not just China but also open-source initiatives led by companies like Meta, Databricks, Mistral, and Hugging Face, according to insights shared by industry professionals with CNBC.
In the previous month, DeepSeek unveiled its R1 model, an open-source reasoning framework claiming to match the efficiency of OpenAI’s o1 model while utilizing less costly and energy-efficient methods.
This breakthrough led to significant declines in market values for Nvidia and other chip manufacturers, fueled by concerns over potential cutbacks in high-performance computing expenditure.
DeepSeek, a newly established AI lab from China aiming for artificial general intelligence (AGI), was founded in 2023 by Liang Wenfeng, who co-founded the AI-oriented hedge fund High-Flyer.
AGI is typically characterized as AI capable of performing cognitive tasks at a level equal to or exceeding human intelligence.
Understanding Open-Source AI
Since the launch of OpenAI’s ChatGPT in November 2022, various researchers have been engaged in examining and enhancing the underlying technologies that propel large language models.
Among their areas of focus is open-source AI, which refers to software whose source code is accessible for modification and redistribution by anyone online.
Numerous entities, from leading tech firms like Meta to innovative startups such as Mistral and Hugging Face, are investing efforts in open-source projects to refine technology and disseminate vital findings with the global research community.
DeepSeek’s Role in Advancing Open Source
The recent achievements by DeepSeek have bolstered the argument in favor of open-source AI solutions, as noted by several industry leaders.
Seena Rejal, the chief commercial officer of AI startup NetMind, remarked that DeepSeek’s success signifies open-source AI’s evolution into a serious, scalable competitor to proprietary models like OpenAI’s GPT.
“DeepSeek R1 has shown that open-source models can reach cutting-edge performance levels, challenging the notion that innovation is the sole domain of closed-source models,” Rejal stated.
Yann LeCun, Meta’s chief AI scientist, echoed this sentiment, highlighting that DeepSeek’s achievements reflect a triumph for open-source AI concepts rather than a straightforward race between nations. Meta is responsible for the widely used open-source AI model known as Llama.
“Viewers who see DeepSeek’s accomplishments and conclude that China is outpacing the U.S. in AI are missing the point. What is truly happening is that open-source models are distancing themselves from proprietary versions,” he commented on LinkedIn.
Explore More on DeepSeek
“DeepSeek has benefited from open research and the support of open source technologies like PyTorch and Llama from Meta. They have innovated by building upon existing ideas and contributions. The availability of their work as open source allows the entire community to reap benefits, showcasing the strength of collaborative research,” he added.
The Global Movement Towards Open-Source AI
With access to advanced chips restricted by U.S. policies, China has increasingly turned to open-source solutions to enhance its AI model capabilities. Firms like DeepSeek are leveraging open-source technologies to foster innovation and broaden the application of their models.
This trend, however, is not exclusive to China. In Europe, a consortium of researchers, companies, and data centers is collaborating on the OpenEuroLLM project—a suite of multilingual large language models.
Jan Hajič, a distinguished computational linguist from Charles University in the Czech Republic, and Peter Sarlin, co-founder of the AI lab Silo AI, which was acquired by U.S. chipmaker AMD, are spearheading this initiative.
This alliance signifies a larger movement towards “AI sovereignty,” where nations strive to cultivate their domestic AI capabilities and infrastructure, aiming to reduce dependency on Silicon Valley innovations.
Challenges Ahead
Despite the promise of open-source AI, experts caution that it may also bring about vulnerabilities, particularly in the realm of cybersecurity. Open-source frameworks can be easily modified and potentially exploited by malicious actors.
Cisco’s recent report highlighted security issues in DeepSeek’s models, finding that R1 harbored critical flaws. Their research revealed that they successfully manipulated R1 to respond affirmatively to a range of harmful prompts, achieving a 100% success rate with such tactics.
“DeepSeek R1 was reportedly developed with significantly lower financial resources than its counterparts, but this raises concerns around safety and security,” wrote Cisco researchers Paul Kassianik and Amin Karbasi.
Additionally, the processing of data through DeepSeek’s R1 can lead to concerns around data privacy, as such information is relayed directly to China. Chinese tech companies have long faced scrutiny over claims that their systems may facilitate state-sponsored surveillance against Western entities.
“DeepSeek poses a dual challenge for businesses and individuals,” noted Matt Cooke, a cybersecurity strategist at Proofpoint. “The innovation potential is significant, but the risks associated with data exposure must not be overlooked.”
He elaborated that sharing sensitive corporate data or personal information with such systems could be detrimental, likening it to granting attackers access to dangerous tools.
Moreover, NetMind’s Rejal reiterated that while open-source AI models push the boundaries of innovation, they also bring cybersecurity concerns that organizations must actively address. This includes risks like software supply chain attacks, prompt jailbreaking, and deliberate attempts to implant unwanted biases or harmful outputs.
WATCH: Why China’s DeepSeek is putting America’s AI lead in jeopardy
Source
www.cnbc.com