DeepSeek, a Chinese AI company, unveiled their v3 design on December 26.
DeepSeek achieved performance that was in line with OpenAI’s GPT-4, a model that reportedly cost over$ 100 million to train, using underpowered chips that were designed to comply with US-imposed restrictions and only US$ 5.6 million in training costs.
Like most Taiwanese laboratories, DeepSeek open-sourced their new concept, allowing everyone to operate their own version of the now state-of-the-art program.
The news came as a result of Silicon Valley’s growing concern that the significant advancement in AI capabilities has now come to an end. It would have appeared that the future of AI lay in marketing and price reduction rather than ability advancements had DeepSeek released their design four times earlier.
Otherwise, the news came a week after OpenAI demonstrated o3, a new model that would position in the 99th percentile of all competing programmers and was properly address the world’s most challenging mathematics problems at 10 times the rate of its predecessor.
Together, the two events point to a new era for the development of AI and a fiercer battle for place dominance between the US and China. China is considerably behind the US in terms of export restrictions tohips, which also fail to handle the next frontier of AI development.
That is the value of reasoning, which teaches AI to believe in the same way as humans do. While earlier models excelled at discussion, o3 demonstrates real problem-solving abilities, excelling not only at tasks that humans find plain, which frequently confounded AI, but also on tests that some Artificial leaders believed were years away from being cracked.
The reasoning method, which is described by Microsoft CEO Satya Nadella as “another scaling law,” could result in improvements like those seen over the past few years thanks to more data and computational power.
Improvements that go along this route are less likely to strain chip capacity limits. Rather, talent, energy efficiency and cheap power will be key.
In Virginia, a major US data center hub, new facilities can wait years just to secure power connections. US utilities and regulators are frantically adapting to the enormous power requirements of advanced AI after 20 years of stagnant demand.
Meanwhile, China is rapidly expanding its power infrastructure, with new integrated computing networks being built across regions like Beijing-Tianjin-Hebei. China’s electricity generation has increased 64 % in the past decade, while the United States ‘ has stalled.
However, just looking at the race from the perspective of a country can be misleading. Instead of releasing the nation’s best model through an established tech company with strong government ties like Tencent, Alibaba, or ByteDance, it was a lab of perhaps 200 people behind DeepSeek and a culture that made the most of that talent.
Although the United States is still a hub for international talent, a recent PNAS article claims that Chinese researchers are leaving the country in greater numbers than ever to study there.
In an interview with the Wall Street Journal late last month, incoming US Secretary of Commerce Gina Raimondo described attempts to halt China as “fool’s errand.”
Ten days later, researchers at China’s Fudan University released a paper claiming to have replicated o1’s method for reasoning, setting the stage for Chinese labs to follow OpenAI’s path.
” The only way to beat China is to stay ahead of them”, Raimondo continued. ” We have to run faster, out innovate them. That’s the way to win”. In the race to lead AI’s next level, that’s never been more clearly the case.
Ben Dubow is the Chief Technology Officer at the AI and intelligence firm Omelas and a senior fellow at the Center for European Policy Analysis.