Recently, exciting advancements in artificial intelligence (AI) have been making headlines. Companies like Perplexity have released new assistants for Android phones, and OpenAI has introduced a capable AI agent that can browse the web to help users. Meanwhile, Google has developed Gemini, which now connects with smart home devices. Meta, known for its social media platforms, is also investing in a huge data center for AI.
However, the real game-changer appears to be a smaller, lesser-known company from China called DeepSeek. They have influenced important aspects of AI, like training costs, access to technology, and overall capability, raising concerns among big players like Google, OpenAI, and Meta.
DeepSeek’s latest model, called V3, was trained for just $5.5 million—far less than the hundreds of millions that larger companies have spent on similar projects. Their CEO, Liang Wenfeng, has built a solid team from top tech firms in China, and the company aims to make AI more affordable and accessible.
Interestingly, DeepSeek’s AI assistant has shown some unusual censorship behavior. When asked about human rights in China, the AI initially provided detailed responses but then retracted them, saying it couldn’t discuss that topic. However, it freely discussed other issues like China’s economy, showing selective filtering of information.
In 2023, DeepSeek launched several new AI models, including one with 671 billion parameters, making it one of the largest generative AI models available. Marc Benioff, CEO of Salesforce, praised DeepSeek, mentioning that its success demonstrates that the real value in AI comes from the data used, not just the technology being sold.
The AI economics are shifting rapidly due to DeepSeek’s much lower costs. For example, their DeepSeek R1 API charges only $0.55 per million input tokens compared to OpenAI, which charges around $15 per million. This change makes AI integration more accessible for many people and businesses.
DeepSeek is also innovating how AI models work. Instead of using the same large amount of resources for every task, their models activate only the parts that are needed, which improves speed and efficiency.
However, questions remain about DeepSeek’s relationship with the Chinese government and how they acquire their technology. There’s speculation about their access to advanced hardware since U.S.-China trade restrictions limit some tech companies’ resources. Still, DeepSeek continues to grow, showing the world that innovation can come from unexpected places
Leave a Reply