Saturday, August 15, 2026

There are several points in today's technology news that are particularly interesting to me. They are connected together and seem to tell a story about technological development, commercial exploration and human nature test.

The first is the peak-to-valley pricing strategy of DeepSeek V4 Pro. Although this strategy can encourage users to use it during free time and reduce the load during peak hours, it also makes many users feel confused and dissatisfied. After all, price fluctuations are no small challenge for users. This made me think that technological progress should not only pursue efficiency, but also consider user acceptance and experience.

Let's look at the internal letter of Want Want Chairman Cai Yanming, bluntly stating that it has encountered a major operating crisis and wants to eliminate employees who have not contributed. Although this approach can improve corporate efficiency in the short term, in the long run, it may damage corporate culture and employee morale. After all, the development of a company is inseparable from the efforts of every employee.

Finally, there was an incident in which an intern nurse at the First Hospital of Shanxi Medical University leaked patient privacy. This incident not only exposed the seriousness of professional ethics and patient privacy protection issues in the medical industry, but also allowed us to see the test side of human nature. In the face of the temptation of interests, some people may forget their responsibilities and moral bottom line.

Although these three events may seem independent, they all remind us that while technology develops, it also needs to pay attention to the following aspects:

The popularization of 1. technology and the user experience after popularization. Technological progress should benefit more people, rather than becoming a privilege for a few. At the same time, user acceptance and experience must also be taken into account.

2. commercialization exploration and humanity test. While pursuing profit maximization, enterprises cannot ignore social responsibility and professional ethics. Otherwise, you may end up paying a higher price.

3. talent training and professional ethics education. The success of an enterprise is inseparable from outstanding talents, and outstanding talents require good professional ethics and professionalism. Therefore, it is crucial to strengthen talent training and professional ethics education.

In short, technology is improving, but we need to be always vigilant and not let technology become a tool for us to pursue interests, but let it truly bring benefits to mankind.

DeepSeek's peak-to-valley pricing and the Token War on AI infrastructure are revealing a deeper trend: the commercialization of AI is moving from an "arms race for computing power" to a stage of "refined operations of efficiency and cost." This is not a simple price increase or promotion, but the entire industry is redefining the boundaries of productivity. The TPW (Token Per Watt) indicator proposed by Shangtang is actually answering a key question: When computing power and power costs become the ceiling of AI development, who can produce more high-quality Tokens with less energy will be able to Take the lead in the next round of competition. This coincides with DeepSeek's peak-valley pricing strategy, which uses price leverage to guide user behavior and divert demand from peak hours to idle hours, thereby improving the overall utilization of infrastructure.

Behind this change is the pressure of AI to transform from "showmanship" to "practicality." The surge in programming capabilities of GLM-5.3 and the positive feedback loop for programming scenarios formed by Grok's acquisition of Cursor indicate that AI competition is shifting from general capabilities to deep optimization of vertical scenarios. This means that in the coming year, we will see more AI companies feeding back real data in specific fields through acquisitions or partnerships to model training, forming a closed loop similar to "data-model-application-data." This model not only improves the performance of models in specific scenarios, but also reduces training costs, because data from real scenarios is often more targeted than general data. Tencent's WorkBuddy and Hunyuan's Co-design strategies are early samples of this trend.

However, this transformation also brings new risks: AI's "situational trap." As models increasingly rely on data from specific scenarios, the problem of "overfitting" may arise, that is, models perform well in specific scenarios, but degrade in general capabilities. More seriously, this model may accelerate the fragmentation of AI, resulting in the incompatibility of AI ecosystems in different scenarios, forming an "App island" similar to the mobile Internet era. In addition, peak-to-valley pricing and efficiency first strategies may make it more difficult for small and medium-sized developers and startups to bear the cost of using AI, further exacerbating the industry's Matthew effect. Those giants with large amounts of user data and scenarios will have a greater advantage in the second half of AI.

Another noteworthy signal is that AI is evolving from a "tool" to an "infrastructure", which means that it will penetrate into various industries like hydropower. Want Want Chairman's internal letter and Honda's shelving of the pure electric vehicle project may seem to have nothing to do with AI, but in fact it reflects the common dilemma of traditional companies in digital transformation: when the market environment changes dramatically, relying on a single product or technology route to "lie and win" The era is over. Want Want's crisis lies in over-reliance on a few flagship products, while Honda's dilemma lies in over-investment in the pure electric route. Both cases remind us that AI, as a new infrastructure, will reshape the competitive landscape of all industries, but only if companies have the ability to "continuously innovate." Companies that cannot adapt to the efficiency improvements and changes in cost structures brought about by AI, no matter how large they are, may face the risk of being eliminated.

From a longer-term perspective, the commercialization of AI is forcing the entire society to rethink the definition of "productivity." In the past, our criterion for measuring productivity was "output per unit of time." In the AI era, the core of productivity may become "intelligent output per unit of energy or cost." This means that future competition is not only a competition for technology, but also a competition for energy efficiency, data quality and scene depth. Companies that can find a balance between efficiency, cost and scene depth will become leaders in the next wave of technology. For ordinary users and developers, how to adapt to this change will be a big challenge.

In recent days, the technology industry has welcomed a number of blockbuster news, including the release of DeepSeek V4 Pro, the launch of GLM-5.3, and the possible sale of Ali Lingxi. These incidents not only reflect the rapid development of current AI technology, but also reveal the competition and cooperation among industry players.

First of all, the release of DeepSeek V4 Pro and GLM-5.3 marks another major advancement in AI technology. These models have significantly improved programming capabilities and security, and can better meet the needs of developers. However, these advances have been accompanied by increasing computing costs and energy consumption. Shangtang Da Device first introduced the concept of TPW (Total Processing Capacity), emphasizing the importance of the efficiency and cost of AI infrastructure. This shows that the future development of AI not only requires more powerful computing power, but also more efficient computing power and lower costs.

Secondly, the news that Ali Lingxi may be sold has triggered people to think about the current situation of China's game industry. The game industry has always been an important part of China's Internet industry, but it has faced severe challenges in recent years. Many giants, including Alibaba, have invested a lot of resources in the game field, but the returns have not been satisfactory. The potential sale of Ali Lingxi may mean the transformation and reshaping of China's game industry, and more innovation and cooperation may be seen in the future.

Finally, recent events have also reminded us of the ethical and security issues of AI technology. For example, the First Hospital of Shanxi Medical University reported that intern nurses violated regulations to disclose patient privacy, warning us of the need to strengthen supervision and security of AI technology. At the same time, the Co-design strategies of companies such as WorkBuddy and Hunyuan also allow us to see the application potential and challenges of AI technology in office scenarios. In general, the current development of AI technology requires not only technological innovation, but also social and ethical thinking and solutions. If these issues and risks are not properly handled, they may limit the further development and application of AI technology.