Why China’s DeepSeek Has US Wowed, And Worried
TOI Explains Chinese-built large language model — DeepSeek-R1 — is significantly cheaper than comparable AI models of Open AI’s ChatGPT or Google Gemini, almost as good at most tasks (in initial checks) and is open for all. Has China levelled the field with US tech giants and even stolen a march on them on the most significant technology of our times?
TIMES NEWS NETWORK: 28.01.2025
In 2023, at an event organised by The Times Of India, OpenAI founder was dismissive about a challenger emerging to something like ChatGPT. When former Google India head Rajan Anandan asked him whether it was possible to come up with a large language model on a relatively low budget, Sam Altman said: “We will explicitly tell you that it’s completely hopeless to challenge us in training foundational models, and you shouldn’t even attempt it.
” Two years later, Altman is probably reassessing what he had said. In a breakthrough that could reshape the future of artificial intelligence, Chinese-built large language model, DeepSeek-R1, is emerging as a cost-effective and open-source rival to advanced AI systems like OpenAI’s o1. Designed to mimic human reasoning, this model processes information step-by-step, making it skilled at tackling complex scientific problems. From chemistry and mathematics to coding, DeepSeek-R1’s early performance, unveiled on Jan 20, has surprised researchers by matching the capabilities of OpenAI’s o1, which had set a high bar for AI innovation just months ago. One of DeepSeek’s most noticeable achievements is its cost-effectiveness.
While industry leaders like OpenAI and Google invest billions annually to develop advanced AI models, DeepSeek version 3 was built for a mere $5.6 million. This stark contrast has turned heads across the AI community, with experts questioning how a relatively modest budget produced results that rival the most advanced AI systems on the market. DeepSeek claims to have outperformed major models, including OpenAI’s GPT-4 and Meta’s LLaMA, on various benchmarks such as maths, coding, and reasoning evaluations. This performance stems from its innovative approach to model training.
By leveraging older-generation Nvidia H-800 GPUs instead of cutting-edge hardware like the H-100 GPUs, DeepSeek sidestepped US semiconductor export restrictions, demonstrating that necessity drives invention. Open-Source Revolution Unlike many proprietary AI models like ChatGPT and Google Gemini, DeepSeek is open source, allowing developers worldwide to access, modify, and build upon its architecture. This openness enables cost effective app development and democratises AI innovation. It is reported that while OpenAI charges $4.40 per million tokens for its API, DeepSeek offers similar services at just 10 cents, making it a highly attractive option for developers. This move toward open-source models could redefine the AI landscape. Historically, once open-source solutions achieve parity with or surpass proprietary models, developers tend to migrate en masse. DeepSeek’s open-source model has the potential to establish China as a leader in the AI space, embedding its technology into global infrastructures and ecosystems. Innovation Under Constraints DeepSeek’s development showcases how innovation flourishes under constraints. US export restrictions on ad vanced semiconductors were intended to slow China’s AI progress. However, these limitations forced Chinese researchers to innovate, resulting in more efficient model training processes.
For example, DeepSeek employed techniques like “distillation”, where a smaller model learns from a larger one, enabling cost and computing efficiency. DeepSeek’s success, once verified and tested more comprehensively, challenges the narrative that US companies maintain an insurmountable lead in AI. Eric Schmidt, former CEO of Google, recently revised his stance, acknowledging that China’s AI capabilities have caught up remarkably in just six months. The rise of open-source models like DeepSeek also poses a strategic dilemma for closed-source giants such as OpenAI and Google. As opensource platforms gain traction, proprietary models may struggle to justify their higher costs. DeepSeek’s emergence signals a broader shift in the AI landscape.
With other Chinese companies, like Kai Fu Lee’s Zero One Dot AI and Alibaba’s Qwen, also achieving cost-effective innovations, the global AI race is becoming increasingly competitive. How Open Can DeepSeek Be? However, DeepSeek’s ascent is not without controversy. Critics point out that opensource models developed in China are subject to govt-mandated “core socialist values”, raising concerns about censorship and the dissemination of biased information. For instance, Chinese AI models have been shown to censor discussions on sensitive topics like Tiananmen Square and human rights abuses. These developments raise critical questions about the future of AI governance. Will the global AI ecosystem adopt democratic, transparent principles, or will it lean toward centralised, state-controlled models? Lessons For India For India and other emerging AI hubs, the rise of DeepSeek offers valuable lessons.
It demonstrates that strategic investments in talent, infrastructure, and efficient methodologies can yield transformative results, even in the face of resource constraints. Clearly, the AI race is accelerating, and the rules of the game are being rewritten.
