DeepMind CEO Critiques DeepSeek's AI Hype

Google DeepMind CEO Demis Hassabis acknowledged DeepSeek's AI model as impressive and potentially the best from China, while also cautioning against overhyping it, noting the model uses existing techniques without groundbreaking scientific advancements

Google DeepMind CEO Demis Hassabis has thrown fuel on the fire by saying that DeepSeek's AI model could be "the best work" to come out of China. However, he quickly toned down the praise by saying that the hype over DeepSeek is "somewhat overstated," and that there are no groundbreaking technological advancements

DeepSeek's Rise

Hangzhou-based DeepSeek was founded in 2023 by Liang Wenfeng, and it rapidly gained attention through its open-source large language models (LLMs). The company released a research paper saying that its AI model was created at a fraction of the cost of its market leaders, utilizing less advanced Nvidia chips. Specifically, the company says it trained R1 for US$6 million compared to $100 million for OpenAI's GPT-4 in 2023, and about one tenth of the computing power used for Meta's comparable model, LLaMA 3. The DeepSeek-R1 model has shown capabilities similar to those of OpenAI's GPT-4o The emergence of DeepSeek's AI model triggered a massive sell-off in U.S. tech stocks and opened up a debate about whether major U.S. technology companies have invested too much in AI infrastructure

Hassabis's View

Hassabis accepted DeepSeek's AI model as "an impressive piece of work," that has "exceptional engineering" which may also "shift dynamics on a geopolitical level" Yet, he noted that the model is not a technological leap forward, as "there are no genuinely new scientific breakthroughs" and that "it is using established AI techniques". He also claimed that Google's Gemini 2.0 Flash models have greater efficiency than DeepSeek's.

Points of controversy

The low cost claims and chips that DeepSeek used have been doubted by experts who believe that the development cost for the models developed by the Chinese company are likely to be more than the declared ones Its censorship compliance by the Chinese government as well as its data collection has also become an issue that attracted regulatory scrutiny in some countries. For instance, Australia banned the use of DeepSeek's technology on all government devices in February 2024

China's AI Ambitions

Despite the debate over DeepSeek's specific achievements, its emergence aligns with China's broader aspirations to become a global leader in AI The company's success has been attributed to government policies, substantial funding, and a robust pipeline of AI graduates. DeepSeek's AI models were developed amid United States sanctions on China and other countries restricting access to chips used to train LLMs