OpenAI CEO Sam Altman recently pointed out that the cost of using artificial intelligence (AI) is falling at an alarming rate, and cited the development of GPT-4 to GPT-4o as an example to emphasize the profound impact of this trend on technology popularization and social economy. The following is a comprehensive analysis:

### 1. AI costs are falling faster than Moore’s Law

- **Descending 10x per year**: Altman made it clear that the cost of using a given level of AI is falling by a factor of 10 approximately every 12 months, and that low prices will drive wider adoption. For example, from GPT-4 in early 2023 to GPT-4o in mid-2024, the cost of processing each token dropped 150 times in about a year and a half, far exceeding the speed of Moore's Law (performance doubling every 18 months).

- **Drivers of cost reduction**: This is mainly due to technological improvements (such as improved model efficiency), hardware innovations (such as dedicated chips) and economies of scale. Altman emphasized that investing more resources (training computing power, data, and inference computing power) can sustainably improve the intelligence level of AI, and the benefits are predictable, in line with the "scaling laws."

### 2. **Technological Development and Economic Impact**

- **Logarithmic relationship between intelligence and resources**: The intelligence of an AI model is logarithmically related to the computing resources and data volume invested, that is, the more investment, the more stable the growth of intelligence. This characteristic has prompted companies to continue to increase their investments. For example, technology giants such as Amazon and Microsoft expect total AI-related capital expenditures to exceed **US$320 billion** in 2025.

- **Super-exponential socio-economic value**: Altman believes that the linear intelligence growth of AI will bring super-exponential economic benefits, such as increased productivity and lower commodity costs (such as in the fields of healthcare and energy), but the prices of scarce resources (such as land) may soar due to increased demand.

### 3. **Future layout of AI Agent**

- **Popularization of virtual colleagues**: OpenAI is working hard to develop "AI Agents" with the goal of making these intelligent agents become virtual assistants in various industries. For example, in the future, agents in the field of software engineering may have the capabilities of junior engineers, although they require human supervision but can handle complex tasks. Altman expects such applications to permeate every aspect of the economy as invisibly as the transistor.

- **Long-term vision of AGI**: Artificial general intelligence (AGI) is seen as the next stage of human innovation, with an impact that could surpass all technological changes in history. Altman proposes that by 2035, each person should be able to mobilize the intellectual resources of the entire human race in 2025 to unleash creativity.

### 4. Potential challenges and policy considerations

- **Risk of unequal distribution**: AI may exacerbate the power imbalance between capital and labor, and policy interventions (such as the allocation of a global “computing budget”) are needed to ensure that the benefits of technology are widely shared.

- **Balance between safety and regulation**: The development of AGI needs to strike a balance between safety controls and individual empowerment to avoid being abused for surveillance or deprivation of autonomy.

### in conclusion

Altman’s perspective reveals a clear path for AI technology from cost reduction to widespread adoption, and highlights its disruptive potential to the global economic structure. However, as technology develops rapidly, society needs to address ethical, distributional and regulatory challenges in advance to achieve the ideal of "AI for all". Investors can focus on long-term opportunities in AI infrastructure (such as computing power and energy) and application scenarios (such as medical care and finance), while being wary of market volatility risks.