Some thoughts and predictions on AI trends

After DeepSeek-R1 became popular during the Spring Festival this year, it can be said that AI was popularized all over the world overnight. This is the most important milestone in the history of AI development since OpenAI launched ChatGPT in 2022. It gives small and medium-sized enterprises and individuals the opportunity to apply AI at low cost and high efficiency. Now that another three months have passed, let me talk about my latest experiences, thoughts and predictions:

1. The development of AI general large models has slowed down in stages. Before the emergence of DeepSeek-R1, AI giants were developing with hidden strength. A year ago, OpenAI caused an internal conflict among the founders of OpenAI because of the controversy over "AI and ethics", which shows that the actual level of OpenAI at that time was already quite high. With DeepSeek directly announcing open source during the Spring Festival this year, all AI giants could not sit still and had to show their cards. This is why we can see top AI technology companies such as OpenAI and Google following closely and quickly launching new versions of better large models. In addition, judging from the recent running scores, the score growth rate of all large models has slowed down, but I think this is a phased correction, and I look forward to the next "DeepSeek moment" of AI.

2. The cost of AI general large models will soon drop significantly. The three core elements of AI - computing power, algorithms, and data - are limited by GPU cards and data sources. After all, these resources are limited, so computing power and data have bottlenecks, and algorithm optimization has become the current breakthrough direction. Just a few days ago, DeepSeek officially issued a message to thank Tencent, because Tencent contributed code to DeepSeek's open source project DeepEP, which can reduce the cost of large model training. Therefore, optimizing algorithms instead of relying purely on hardware is the next trend, and the natural benefit is to reduce the cost of applying AI.

3. The general AI model will be reshuffled within the year, and only a few players will remain. Computing power, algorithms, and data require strong capital support. Just like the cloud computing and public cloud wars of the year, only a few players will remain in the end. The same will be true for AI. The game of big fish eating small fish has begun. In the current economic environment, this game may end early.

4. Small AI models in vertical fields have little chance of independent development, but great opportunities for AI applications. I will still draw on the analogy of the development of public clouds. Those vertical IAAS/PAAS and even SAAS companies are all backed by cloud giants. They are either strategically invested or directly acquired, and it is difficult for them to develop independently. At present, from the perspective of public opinion, the idea of ​​small AI models is indeed fading. Instead of talking about models, people pay more attention to applications, including capital has also turned to AI applications. The main reasons behind this are: first, people's overall understanding of AI has improved; second, the data advantage barrier of small models is not strong; third, companies must face the reality of capital and commercialization.

5. Many people pay attention to AI, but few actually study and invest in it. Whether it is a company or an individual, based on my own experience and that of those around me, I feel that people are still relatively ignorant and confused about AI. I think there are several reasons for this: First, AI is still relatively technical, with many concepts and terms, and it is difficult to systematically understand it through fragmented information on self-media; second, the evolution and iteration speed of AI is too fast. In summary, the threshold for learning and research is too high.

6. AI should not be seen as a trend, but as a trend. The trend will soon pass, and the result of chasing the trend may be nothing, but the trend exists for a long time. For AI, everyone is now at the same starting line. Even if you stand out a little bit, you will surpass your peers. Here is a reminder to yourself to strengthen your study of AI.

7. Strengthening of policies on AI supervision. Some AI-related policies have been issued. For example, AI-generated content must be marked. In the future, AI applications may need to be registered and qualified, similar to website registration and licenses in some industries, because registration and qualifications are the main means of supervision. This is true in all industries, including finance and K12 education and training, so AI will definitely continue to be so.

8. SMEs should be pragmatic about AI. Return to the original intention, mission, and vision of your own business, return to the pain points, needs, and value of the users you serve, and think about a few questions: What problems can AI help you solve for users and what value can it create? Why should users pay for it? Don’t blindly use AI for the sake of AI. This is the real feeling in practice.