Arthur Goldstuck | CEO | World Wide Worx | Editor-in-chief | Gadget.co.za | mail me |
In “Seven Days That Changed the World”, a small Chinese startup made a rapid and remarkable shift that splits the Artificial Intelligence (AI) atom. Last month, it went from quietly releasing a new AI chatbot to fending off cyberattacks, security leaks, accusations of theft and allegations of sanctions-busting.
DeepSeek-R1 formally launched on January 20. This event was comparable in significance to the release of ChatGPT on November 30, 2022. However, its full implications took time to reach the mass media. One reason for this delay was the media’s lack of technology literacy. Few understood the implications of a new AI model with reasoning capabilities similar to OpenAI’s O1, the most advanced version of ChatGPT available to the public.
Reaction to the DeepSeek breakthrough
A week later, news spread that DeepSeek, a 20-month-old company, had developed the model at a fraction of the cost of leading chatbots.
Reactions ranged from delight to disbelief to desperation. These reactions offer insights into DeepSeek-R1. For one, it was not only a highly cost-effective model to develop but also promises to be economical for its users and developers who want to integrate it into their tools.
DeepSeek-R1 was released as open-source software. This allows anyone to adapt it for their own purposes. Using more advanced versions or reducing restrictions also costs substantially less than established AI platforms. Former Intel CEO Pat Gelsinger publicly declared he would adopt it for his church-oriented startup because of its cost-effectiveness.
DeepSeek splits the AI atom as the response from current AI leaders has been one of disbelief. Many refuse to accept DeepSeek’s detailed reports about how it trained its models. The startup claimed it developed several innovations. One of these innovations was teaching an AI model how to reason without human supervision. It also revealed a four-step approach that progressively treated the model as a baby, child, adult and mature adult. Researchers on the ground have applauded DeepSeek’s innovations.
Mark Chen, chief research officer at OpenAI, congratulated the company, saying:
Congrats to DeepSeek on producing an O1-level reasoning model! Their research paper demonstrates that they’ve independently found some of the core ideas we did on our way to O1.
However, investors in OpenAI were not as optimistic. Neal Khosla, son of legendary venture capitalist Vinod Khosla, who invested $50 million in OpenAI, acted as a proxy for the investor community.
Khosla criticised DeepSeek on X, claiming that:
DeepSeek is a CCP state psyop + economic warfare to make American AI unprofitable” (sic). They are faking the low cost to justify setting the price low, hoping everyone switches to it, damaging AI competitiveness in the US.
Accusations of theft without evidence
The term “psyop” or psychological warfare, seemed to be the last resort of an American conspiracy theorist, especially since no evidence was provided.
OpenAI itself fueled this narrative by claiming to have evidence that DeepSeek had stolen its work to train its models cheaply. However, no such evidence was presented. The accusation felt rich, especially since ChatGPT was trained by scraping content from websites – essentially “stealing” almost all text available on the internet.
Disparity in development costs
Behind the scenes, the true anger was likely directed at the stark financial disparity between DeepSeek and its competitors.
DeepSeek-R1 cost just $5.6 million to build, while OpenAI spent $6 billion developing ChatGPT. This disparity must have caused anxiety among the venture capital firms, wealthy investors and sovereign funds who have invested billions into AI startups. They may have started to fear that their money was being wasted.
In February, Stanford University researchers spent only $50 training a model from Chinese giant Alibaba to the same level as O1. It’s no surprise that investors are looking for scapegoats.
At present, the Chinese Communist Party (CCP) has become an easy target. This reaction, however, was not the most absurd. The most absurd came from ordinary share investors who adopted their usual herd mentality. They sold off any stock with a hint of AI. Nvidia, which had been the darling of investors for the past year due to its AI-optimized chips, saw a massive 17% drop. By mid-week, it had clawed back a quarter of this drop. For those who understood the relationship between cheaper AI, more demand for AI, and more demand for chips, this probably represented a buying opportunity.
China’s rise in AI research
The biggest surprise was that DeepSeek came as such a surprise in the first place. Many Americans were shocked that a Chinese company had caught up so quickly. For years, research has shown that China has been catching up with the US in high-quality academic research papers on AI.
One metric for measuring the quality of academic papers is their share of citations in other works. On this front, China surpassed the US in 2020, according to the Stanford University’s AI Research Index, as reported in its 2021 edition. However, to ensure quality, one must measure the citations of the most influential papers.
As early as 2019, an analysis by Semantic Scholar of over two million AI papers showed that China was poised to overtake the US in the most-cited 20% of papers by 2020. By 2025, China was expected to surpass the US in the 1% of most-cited papers. It appears that 2025 arrived much sooner than expected for the American AI industry.
