Tuesday, June 30, 2026

Recently, the AI circle has been really ice and fire. On the one hand, Jack Clark, co-founder of Anthropic, was boasting that the 2028 RSI recursive self-improvement would come true, and AI would have to build its own ASI; on the other hand, Lenovo created a big model to predict the World Cup, but in the end, when it came to a draw, it would be blinded. Only 11 in 20 games. To put it bluntly, large models are still fortune-tellers with excessive computing power when handling highly uncertain game scenarios. If we are engaged in engineering, don't be distracted by this grand narrative. It depends on what real problems it can solve.

In fact, from the perspective of business trends, everyone has already begun to calculate ROI. China Mobile has set up a Token office. This signal is very clear. The inside of the big factory has begun to use Token as water, electricity and gas for resource allocation and settlement. In the past, each department managed and allocated GPUs, but now it uses Token transfer to get through. This is a sign that project management is mature. Looking at Wal-Mart, people don't play tricks with you, and directly turn 5000 offline stores into AI infrastructure for agency shopping. As long as traditional giants are not stupid, the physical assets in their hands are moats. If a startup wants to subvert others by putting a GPT shell, it is purely overthinking. The same is true for financial AI. Want to use AI to predict that the market will cut off leeks? The one-vote veto power of supervision teaches you how to behave in minutes. Compliance is the biggest moat. Don't use users 'wedding money to test the illusion of models.

For us ordinary developers, don't stare at those lists of hundreds of billions of parameters all day long. You see, foreigners on HN are praising Qwen 3.6 27B as a locally developed sweet spot, that's right. A good model is a model that can run locally and can truly solve code completion and document processing. The Unlimited OCR of Baidu swipes the list on HF. The same goes for the same reason. If the dirty work of OCR is done well, it has more commercial value than any general model. The person in charge of OpenAI Codex said quite thoroughly that now the cost of code implementation is close to zero, the order of product making is reversed, and aesthetics and judgment are scarce resources. In the future, everyone will fight not about who can type out the code, but who knows what to do and how to do it to make it work. So, don't worry about the arrival of ASI, straighten out the Agent workflow at hand, and reduce the cost of Token, which is better than anything else.

Among today's technological signals, earthquakes, sports event disputes and the rapid development of AI technology have become the focus. These events not only reflect the current status of social and technological development, but also predict possible trends and challenges in the future.

The earthquake in Yibin, Sichuan once again reminds us that natural disasters are inevitable challenges in the development of human society. The earthquake not only affected the lives of local residents, but also had an impact on infrastructure and social order. In the process of responding to natural disasters, the role of science and technology has become increasingly prominent. For example, earthquake early warning systems, drone rescue and artificial intelligence analysis of earthquake data can greatly improve rescue efficiency and reduce casualties.

At the same time, controversial events in sports events, such as the dispute between Algeria and Austria and South Korea's president's criticism of the national football team, reflect the contradictions in the commercialization and socialization process of sports. On the one hand, the commercialization of sports events promotes the development of the sports industry and increases the income of athletes and clubs; on the other hand, commercialization has also brought about disputes in aspects such as profit transfer and fair competition. These disputes not only affect the fairness of sports events, but also trigger social reflection on the spirit and values of sports.

In the field of AI technology, the countdown to RSI (Recursive Self-Improvement) is accelerating, indicating that AI will enter a new stage of development. AI is no longer just a tool, but has begun to evolve independently, which will have a profound impact on all walks of life. For example, in the financial sector, the emergence of AI stock trading tools has changed traditional investment methods, but it has also brought risks and regulatory challenges. In the field of product development, the application of AI has changed the traditional development process, making aesthetic judgment the most scarce resource.

The impact, value, problems, and risks of these incidents are multi-faceted. First of all, responding to natural disasters requires the support of science and technology, which requires us to strengthen scientific and technological research and development and infrastructure construction. Secondly, disputes over sports events require us to reflect on the commercialization and socialization process of sports and find a balance point. Finally, the development of AI technology requires us to pay attention to the ethical and social issues it brings to ensure that the development of technology can benefit mankind.

In short, today's technological signals reflect the complexity and variability of social and technological development. In future development, we need to pay more attention to the interaction between science and technology and society, use the power of science and technology to respond to challenges, and at the same time be vigilant about the risks that technology may bring.

Anthropic executives predict that AI will achieve recursive self-improvement in 2028, as if the singularity is close at hand, but reality is much more bony than PPT. AI predicts that its true colors will be revealed once the World Cup is drawn, and financial AI stocks will make retail investors lose their money. When the industry is raving about the cost of code generation reaching zero, no one is investigating who is responsible for AI's wrong decisions. Every time there is a technology bubble, shovel sellers selling computing power always make a lot of money, and retail investors and consumers who blindly follow suit are the final ones who pay the bill. Traditional giants are eager to set up Token offices. This kind of follow-up amidst anxiety about computing power is still far from a closed loop of business.

The cycle of hardware technology is equally cruel. BOE burned out 500 billion yuan in capital expenditure and waited for more than 20 years to let shareholders see the money back, revealing the truth about the heavy asset manufacturing industry. However, capital is always forgetful, and perovskite space photovoltaics are clamoring to reduce the cost to one-tenth and mass production next year. There are often countless pits worth 500 billion yuan between laboratory formula optimization and commercial mass production. When storage giants are sued over price monopoly, we should be even more vigilant about new technologies that use the banner of disruptive costs. The space economy is a good story, but before large-scale benefits are generated, grand narratives are mostly to make money in the primary market.

Technological evolution has never been a linear and upward movement. Whether it is the countdown to self-improvement or the commitment to mass production of space photovoltaic, they are using future visions to cover up current business flaws. The real moat is not how many Tokens it consumes, but whether it can survive strong cycles and strict supervision. Don't let the countdown bell rush you to pay your IQ tax. It is much more important to see who is making money and who is paying the bill than chasing the wind.