DeepSeek, a mobile AI application developed by a Chinese startup, has climbed to the top of the iPhone free app charts in both China and the United States, overtaking OpenAI's ChatGPT. This remarkable achievement comes shortly after the launch of DeepSeek's new \"reasoning model,\" DeepSeek R1.
Enhancing AI Reasoning with DeepSeek R1
When users activate the \"DeepThink (R1)\" mode within the app, they can witness the AI's \"thinking process\" as it formulates responses. This feature allows the application to tackle complex logical and mathematical problems more effectively. According to DeepSeek's official website, the R1 model's performance is \"on par with\" OpenAI-o1, while operating at approximately one-thirtieth of its rival's cost.
The mobile app and web chatbot are currently free for general use, with only API calls for programmers requiring payment. Additionally, DeepSeek has made its full-size model available for free download, enabling users with adequate hardware to run it locally. For those with less powerful devices, the company offers scaled-down versions of the model, optimized for everything from ultra-thin laptops to high-performance gaming rigs.
Beyond providing free access to its models, DeepSeek has also published a research paper detailing the development of R1, allowing other developers to replicate the process using their own training data.
Resonating Impact in the AI Community
The release of the R1 model has created waves in the AI industry, drawing impressed reactions from leading figures in the field.
Marc Andreessen, venture capitalist and co-founder of Netscape, described DeepSeek R1 on X.com as \"one of the most amazing and impressive breakthroughs\" he has ever seen, calling it \"a profound gift to the world.\"
Perplexity CEO Aravind Srinivas noted on X.com that \"DeepSeek has largely replicated OpenAI-o1-mini and has open-sourced it.\"
Jim Fan, a senior research manager at Nvidia, praised DeepSeek as a \"non-U.S. company\" upholding OpenAI's original mission of conducting \"truly open, frontier research that empowers all.\"
Yann LeCun, Meta's chief AI scientist, emphasized that DeepSeek's success does not signify \"China surpassing the U.S. in AI,\" but rather highlights how \"open-source models are surpassing proprietary ones.\"
Kai-Fu Lee, former president of Google China, expressed a sense of vindication, stating that the DeepSeek releases validate his belief in China's potential to excel in generative AI engineering, despite skepticism from others.
DeepSeek CEO Advocates for Chinese Innovation
In a July 2024 interview with 36kr.com, DeepSeek CEO Liang Wenfeng argued that China must move beyond the stereotype of merely applying U.S. innovations. As the country's economy grows, he said, it should transition into a global contributor to innovation rather than relying on the advancements of others.
Liang emphasized that innovation stems not only from the pursuit of business success but also from genuine curiosity. He revealed that the DeepSeek team is composed of young talent and that the development of the company's earlier V2 model did not involve any overseas Chinese contributors.
\"Perhaps the top 50 talents in this field are not in China,\" Liang told 36kr.com. \"But we can cultivate our own.\"
Reference(s):
cgtn.com