Chinese AI DeepSeek Surpasses ChatGPT in App Store Rankings

On Monday, the artificial intelligence (AI) assistant of Chinese startup DeepSeek overtook its popular rival ChatGPT to become the top-rated free app on Apple's App Store in the United States.

Since its January 10 launch, US users have become interested in DeepSeek-V3. According to its developers, it "takes place on the leaderboard among open-source models and competes against the world's most advanced closed-source models."

The benchmark reflects how DeepSeek has left a deep impression in Silicon Valley. It challenges the widely held views that the United States has a premier position in AI and that Washington's export controls against China would have some gating effect on advanced chip and AI capabilities. 

Innovation in the Face of Limitations

Given the limitations placed on electronic exports from the United States, DeepSeek's success is even more impressive. This is because DeepSeek makes it feasible for a nation like China to produce high-power chips like Nvidia's H100. Nevertheless, DeepSeek trained its models using older chips from Nvidia's A100 stockpile, which had been purchased alongside lower-capacity H800 chips prior to the sanctions.

DeepSeek engineers have been looking for ways to optimize their training exercises to use less memory and processing power, circumventing those hardware limitations. Their focus on efficiency has allowed them to compete on performance against systems built with far more significant resources.

"The US export control has essentially cornered Chinese companies into tremendous resourcefulness," said Matt Sheehan. He is an AI researcher at the Carnegie Endowment for International Peace. "It has incentivized innovation."

Global Recognition of Deepseek

Microsoft's CEO spoke positively about DeepSeek's accomplishments at the World Economic Forum in Davos. "It is impressive how well they have built a compute-efficient, open-source model. Developments like DeepSeek's should be taken very seriously," he expounded. This model solves problems step-by-step using a "chain of thought" reasoning technique similar to ChatGPT. By focusing on getting it right rather than providing every logical step, time is saved, and quality is maintained. Dimitris Papailiopoulos praised the model for its simplicity. DeepSeek was very focused on R1's actual core performance.

Read Also: The Deportation Dispute Between the United States and Colombia: Trade Sanctions and Diplomatic Agreement