DeepSeek: The New Kid On The AI Playground

A new AI has emerged, capturing global attention and making waves across industries. DeepSeek, a China-developed chatbot, is proving to be more than just another artificial intelligence: it’s a disruptor, rattling global markets and reigniting the U.S.-China tech rivalry. Built at a fraction of the cost of its Western competitors, it symbolises China’s growing ambition to lead the global tech race, challenging long-held assumptions that cutting-edge innovation is exclusive to the West. Whether DeepSeek enjoys short-term success or marks the beginning of China’s tech-driven future, one thing is certain: the landscape of AI competition has shifted.

Beyond being another generative AI-powered chatbot akin to OpenAI’s ChatGPT or Google’s Gemini, DeepSeek is designed for a range of tasks, including coding, mathematics, content creation, and human-like reasoning, all while operating at significantly lower computational costs than its competitors. The mastermind behind DeepSeek, Liang Wenfeng, revealed that training the model cost just $6 million. This was achieved by leveraging a combination of advanced Nvidia A100 chips and less sophisticated alternatives. DeepSeek’s training methodology further sets it apart, leveraging four key techniques: learning by trial and error (reinforcement learning), a custom reward system to guide learning (reward engineering), making it smaller while keeping its intelligence intact (distillation) and, lastly, recognising complex patterns on its own just by practicing, without being directly programmed to do so (emergent behaviour). 

DeepSeek’s rapid success and its ability to rival, and in many cases outperform, Western models has raised significant questions about how companies approach AI innovation. Few anticipated that a Chinese company would be able to leverage existing resources in such unconventional ways or that DeepSeek’s advancements would have had a profound impact on the global chip industry. It has now been proven that state-of-the-art AI performance can be achieved with fewer advanced chips, a revelation that has already sparked concerns at major companies about the future demand for high-performance chips. Nvidia suffered a historic one-day loss in market value of almost $600 billion. Additionally, unlike its competitors, DeepSeek is free and open source, further intensifying its disruptive influence. 

On a global scale, DeepSeek’s breakthrough highlights China’s growing technological independence as it seeks to reduce reliance on Western-dominated tech ecosystems and position itself as a leader in high-tech industries. It also challenges the assumption that innovation is tied to enormous budgets and cutting-edge chips used by American companies, instead demonstrating that proper research and a passionate, dedicated team of developers, regardless of experience, can drive significant innovation. The cost-effective nature of DeepSeek’s model also contributes to the increasing importance of sustainability in AI. As fossil fuels dwindle, the demand for energy-efficient, low-carbon AI models will only intensify and DeepSeek’s success with limited resources has paved the way for a more environmentally conscious approach to AI development. 

While DeepSeek’s rapid rise has been impressive, it has not been without its share of controversies. The  political censorship in line with China’s regulations has raised concerns about the ethics of AI in different political systems, as well as the objectivity and reliability of AI models developed under authoritarian regimes. While DeepSeek may comply with government policies, it also risks perpetuating biased or incomplete information. Currently, users can bypass these restrictions by phrasing questions in a specific way (such as replacing every “i” with “1” and “o” with “0”) but this workaround represents a limitation that could hinder the model’s broader adoption, particularly outside China, where users expect AI to provide unfiltered, unbiased information. Privacy concerns also loom large, with critics warning that DeepSeek could be vulnerable to government influence, potentially jeopardising user data while on the security front, recent large-scale cyberattacks and the subsequent temporary suspension of new registrations have exposed several system vulnerabilities.

Liang was clear about the vision behind DeepSeek: “China’s AI cannot remain a follower forever.” The project aspires to be a symbol of China’s ability to lead the global AI race, going beyond commercial success and bridging the gap between imitation and true innovation. While its rise has certainly shaken up the AI industry, its long-term ability to compete with well-funded Western models like OpenAI is still uncertain. Whatever the case, China’s AI trajectory has challenged long-standing assumptions about the supremacy of American-led innovation and is signalling a gradual shift toward a more multipolar technological world.

Image: Flickr/Tim Reckmann

Licence.

No image changes made.