सिंहदरबार संवाददाता
Kathmandu. US export controls on high-tech chips have inadvertently boosted the success of startup DeepSik’s AI chatbot. It has raised fears that Washington will do nothing to stop China’s push for global dominance in AI.
The company, based in the eastern Chinese city of Hangzhou, has surprised investors and industry experts with its R-1 program. It can compete with American competitors at a significantly lower cost. This is despite strict US rules that prevent Chinese companies from accessing advanced chips needed to run the massive learning models used to develop AI.
DeepSik founder Liang Wenfeng has admitted that the “restriction on high-end chips” has been a major obstacle to its work. But analysts say the restrictions, while aimed at ensuring American technological dominance, may have spurred the company to develop clever ways to circumvent them. The company said it used less-advanced H800 chips to run its big learning models. The chips were allowed to be exported to China until the end of 2023.
“The restrictions on China’s access to the chips forced the DeepSik team to train more efficient models that could compete without the huge computer training costs,” said Jeffrey Ding of George Washington University. DeepSik’s success, he said, “shows that US export controls are ineffective at preventing other countries from developing leading-edge models.”
“History tells us that it is impossible to bottle up general-purpose technologies like artificial intelligence (AI),” he said. DeepSik is not the first Chinese company to be forced to innovate in this way. Tech giant Huawei has returned to profitability in recent years after restructuring its business to deal with US sanctions. But it is the first to cause such panic in Silicon Valley and Washington.
Venture capitalist Marc Andreessen has described it as a “Sputnik moment.” Sputnik is a reference to the Soviet satellite launch that exposed the vast technology gap between the United States and its major geopolitical rival. – Part of the cost – For years, many took American dominance in AI for granted. The field was dominated by big Silicon Valley names like OpenAI and Facebook-parent company Meta. While China has invested millions and pledged to become a world leader in AI technology by 2030, its efforts have not been enough to raise concerns across the Pacific.
Tech giant Baidu’s attempt to compete with ChatGPT failed to impress when its Arnie bot went public. It seemed to confirm many people’s view that Beijing’s strict regulatory environment for big tech companies would stifle any real innovation. That was coupled with tighter regulation under Joe Biden’s administration. Its aim was to limit Chinese purchases of high-tech chips needed to run AI big language models. But DeepSec has overturned many of those assumptions.
“It overturns long-held assumptions that many people have about the computing power, the data processing, that are needed for innovation,” said Sam Sachs, a law research scholar and senior fellow at Yale Law School’s Paul Tsai China Center. “So the question is, can we get cutting-edge AI at a fraction of the cost and a fraction of the computation?” DeepSec’s model emphasizes cost reduction and efficiency, while U.S. policy toward AI has long been based on assumptions about scale. According to Ding of George Washington University, “You add more and more computing power and performance to the problem to get better and better performance.”
That’s the central idea behind President Donald Trump’s Stargate Ventures, a $500 billion initiative led by Japanese giant SoftBank and ChatGPT maker OpenAI to build the infrastructure for artificial intelligence. But the success of DeepSik’s R-1 chatbot suggests that innovation can come much cheaper.
DeepSik’s developers claim to have built it for just $5.6 million. Some urge caution, arguing that the company’s cost-saving measures may not be so innovative. “DeepSec V3’s training costs are competitive, but they are within historical efficiency trends,” said Lennart Heim, an associate information scientist at the Rand Corporation, referring to the previous version of R-1. “AI models have been getting cheaper to train over time – this is not new, we don’t even see the full cost picture of infrastructure, research and development.” – Wake-up call – However, Trump has described DeepSec as a “wake-up call” for Silicon Valley, saying they need to be “laser-focused on competing to win.”
Former US Representative Mark Kennedy said DeepSec’s success “does not undermine the effectiveness of future export controls.” He said Washington could take further steps by “expanding restrictions on AI chips” and increasing scrutiny of what technology Chinese companies can access. But Kennedy, who is now director of the Wilson Center’s Wahba Institute for Strategic Competition, said it could also strengthen its own industry.
“Given the limitations of purely defensive measures, it could refine policies to increase domestic AI investment, strengthen alliances, and ensure China maintains its leadership in the AI ecosystem without inadvertently pushing more nations into it,” he said. “The real fear of falling behind China could now motivate that push,” said Rebecca Arcesati, an analyst at the Mercator Institute for China Studies (MERICS).
प्रतिक्रिया दिनुहोस्