2025 a crucial year for AI applications

The Government Work Report on Wednesday highlighted new achievements including integrated circuits, artificial intelligence (AI), quantum technology, and other areas.

In fact, a new wave of AI technology, led by large language models, has been sweeping across the globe since 2023, prompting many countries to make AI development part of their national strategy, with the United States adopting the "All in AI" strategy to dominate the global AI industry. China, too, has expedited its AI development, launching the "AI plus" initiative and implementing policies to make AI a core engine of development for new quality productive forces.

Previously, China's large AI models had certain limitations in terms of computing power and data. But innovators such as DeepSeek have recently made significant progress in AI reasoning through reinforcement learning, pioneering a new direction for AI. By embracing an open-source approach, Chinese innovators have broken the technological barriers of leading models, accelerating the commercialization of large models. This breakthrough is not only a technological leap for Chinese AI companies vis-a-vis US giants such as OpenAI, it also narrows the gap in AI capabilities, dispelling concerns over chip limitations and showing China's ability to develop a global technology paradigm.

With the widespread adoption of AI applications such as DeepSeek-R1 and Nano AI Search, smart technology including AI is gradually becoming accessible to all. Also, an increasing number of local governments and businesses, including small and medium-sized enterprises, are adopting large model systems to accelerate industrial applications and digital transformation.

The global competition for large models has entered a critical stage, and to gain an advantage in this competition, China should continue to achieve more breakthroughs in core technologies and their applications. Thanks to its diverse industries and development environments, China has a unique advantage in using large models to power a new industrial revolution. I believe 2025 will be the year of AI applications, with the key to success lying in seizing the opportunities created by DeepSeek, and combining local advantages with different application scenarios to stay ahead in the global AI competition.

DeepSeek has triggered an application explosion by making industries more efficient, reducing costs and reconstructing products. In fact, startups empowered by DeepSeek are seizing the new opportunities, creating an "iPhone moment" for industries. Based on DeepSeek's capabilities, breakthroughs can be achieved in six fields.

The first application of large models will lead to "universal intelligence" for all. Given DeepSeek's open-source and low-cost features, individuals can now own large models and become "super individuals".

The second is "intelligent everything". As AI transforms smart hardware, and large models shrink in size, AI devices — from smart-connected cars to personal computers and mobile phones — will become the standard. And with AI's help, the scope of smart hardware application will also expand.

The third is the empowerment of traditional industries to "digitally transform and innovate". Large models' multi-modal capabilities can process and generate data from images, videos and audio clips or pieces, transforming unstructured data into knowledge. This will prepare the ground for large models in business applications, boost core competitiveness and expedite the digital transformation of traditional industries.

The fourth is "future industries". With lower costs and reduced computing power requirements, large models can be used in intelligent devices like robots. For instance, predicting driving sequences for fully autonomous cars or the movement of robots enhances embodied intelligence.

The fifth is "scientific research". DeepSeek's strong inference model can help users leverage professional knowledge in scientific fields for reinforcement learning, creating specialized scientific models. Large models are changing how fundamental scientific problems are studied, and addressing issues such as energy development.

The sixth is "AI security". As DeepSeek expands, challenges like hallucination and prompt injection attacks arise. Injection attacks occur when attackers exploit weaknesses in how software or systems handle user input prompts. By cleverly crafting malicious input, they trick the system into performing unintended actions, with the goal of accessing sensitive information, disrupting operations, or gaining unauthorized entry. To ensure the application of large models in enterprises, they need to be linked to specialized knowledge databases and IT systems, which in turn could increase security risks.

However, to deal with such issues, 360 Security Group has introduced a new solution-"model-driven security" — to address potential security issues arising out of large model applications. The 360 group also has formed a large model security alliance, which includes industry partners, research institutions and end-users, to resolve security issues and build a collaborative model for the AI security industry.

The author is founder and chairman of Chinese cybersecurity company 360 Security Group, and a member of the 14th National Committee of the Chinese People's Political Consultative Conference. The views don't necessarily represent those of China Daily.

If you have a specific expertise, or would like to share your thought about our stories, then send us your writings at opinion@chinadaily.com.cn, and comment@chinadaily.com.cn.