DeepSec’s focus on low-cost text processing has caught the attention of the artificial intelligence industry. AI innovation has a more promising future as China and the US work together and compete. Technologist Yin Ruozhi looks at the benefits of a deeper security strategy to win the competition.
DeepSec’s rapid growth and global reach raises an interesting question: Why has it captured the AI industry in China and the US and inspired so many startups to build their businesses around it?
The solution lies in reducing costs and continuing to focus on text processing rather than features like text-to-image generation, which OpenAI is known for. I previously highlighted the fact that the relatively low cost of in-depth study attracted the attention of the entire industry in another article in this series, to be published in 2024.
A year ago, a consensus emerged among Chinese AI scientists that the primary focus of this wave of large-scale language modeling (LLM) technologies would be high-quality text analysis, summary, and parsing, not content creation. The term “creative artificial intelligence”, which is often used in the media, is therefore very confusing for people outside the field of artificial intelligence.
If users look at AI primarily in a creative way, they may lack the judgment and intervention skills needed to evaluate the issues being raised.
An important feature of this generation of large linear models (LLMs) is that analysis often requires a good understanding of the results and associated constraints. LLM graduates can sometimes “think”, meaning they produce content that is misleading. When this happens, the user should engage in controlling and improving the result by performing multiple iterations to get the best possible result.
If users look at AI primarily in a creative way, they may lack the judgment and intervention skills needed to evaluate the issues being raised. This gap is why personal media often achieve impressive results thanks to artificial intelligence, while the average consumer only experiences mediocre results in areas like music or video creation.
One key challenge is the inability of users to critically think about and fine-tune AI results. If the tasks involve analyzing, summarizing, and evaluating text and the user has strong feelings about the content, design can be more easily implemented. For example, a teacher can use artificial intelligence to assess students’ broad work and refer it for detailed assessment and assessment of the student’s understanding of key concepts. Similarly, a financial analyst can use extensive research methods to conduct extensive market research and actually derive valuable insights.
The current wave of LLM programs benefits China’s e-commerce industry. E-commerce product owners need to process a wealth of data, including user exposure, competitor sales data, and third-party marketing strategies, and then analyze and summarize it to shape their product strategy.
For example, the product owner of a good company faces competition from 37,000 such products on popular e-commerce platforms such as Taobao, JD.com, and Pinduoduo. Before the advent of the LLM (Master Learning Law), market analysis often required an entire team. Beyond verifying information themselves, product teams rely on information they can’t process themselves. Now that’s it
The problem is that if that part of the family doesn’t take good care of these items or makes a mistake, important information could be lost.
In the food and entertainment e-commerce projects I work on, such failures can lead directly to failed marketing and product update plans, with a probability of more than 40%. With the emergence of advanced artificial intelligence models, the product manager no longer needs an assistant. You just need to collect the target data and let the main model read it “honestly” and respond according to the user’s requirements.
At present, the Chinese e-commerce sector is the most active and suitable sector for the application of a comprehensive set of AI modeling technologies.
As a product manager, he can easily evaluate the results by issuing continuous improvement orders to technical staff or directly reading the source data. Therefore, in such application scenarios, large AI models can significantly improve performance without drawbacks such as LLM illusions affecting the results.
One of the main reasons why DeepSeek prioritizes reducing the cost of content creation, such as graphic text generation, is that the first wave of real-world applications of large models were in content analysis, summarization, and text analysis – areas where cost sensitivity is critical.
China is currently the most developed country in the world in terms of the e-commerce industry (even the US e-commerce industry is not as developed as China). At present, the Chinese e-commerce sector is the most active and suitable sector for the application of a comprehensive set of AI modeling technologies.
In the field of artificial intelligence, there is a huge exchange and complementarity between China and the United States in terms of talent and technology – and the information gap is small.
The interesting question is also whether our American friends are aware of this? The answer is: Yes. In the field of artificial intelligence, there is a huge exchange and complementarity between China and the United States in terms of talent and technology – and the information gap is small.
Two factors may indicate this. After the US election, Elon Musk, the richest man in the world and head of the Department of Productivity, conducted a comprehensive audit of spending across the US government using artificial intelligence technology and identified areas of foreign spending. The purpose of this application is primarily to perform high-quality analysis, summarization and verification of large-scale information.
OpenAI is now fully aware of this trend. After the overwhelming success of DeepSeek, the most striking reaction to OpenAI among industry players was the release of Deep Research. Providing in-depth analytical reports to professionals in fields such as finance, science, and engineering; For example, financial analysts can use it to assess supply chain risks in a specific industry.
In the latest Humanities test, the Deep Research model achieved 26.6% accuracy on expert-level questions, breaking previous records. It also performed well on the GAIA test, achieving high accuracy across a range of difficulty levels, especially on Level 3 tasks that require complex, multi-step search and synthesis. This application area also includes analyzing, summarizing, and validating complex, high-quality information.
