OpenAI CEO reaffirms support for the scaling law, which has been pointed out as outdated, saying, 'If AGI is realized, many products will become extremely cheap,' and 'The economic value of AI will continue to increase, so we should continue to invest in it.'

OpenAI CEO Sam Altman updated his blog and talked about the impact that AGI (artificial general intelligence) will have on the world and three insights about the economics of AI that have been obtained so far. According to him, the invention of AGI will be a major revolutionary point for civilized society on par with the invention of electricity and the Internet, and many products will become available at significantly lower prices due to the reduction in product development costs.

Three Observations - Sam Altman
https://blog.samaltman.com/three-observations

In his blog, Altman states that 'we are beginning to see systems that can be called AGI,' and lists the following three observations about the economics of AI:

1: AI models become more intelligent as the amount of computing power spent on training, the amount of learning data used for training, and the amount of computing power spent on inference increase. It is possible to obtain continuous and predictable benefits from the costs spent, and the scaling laws used to make this prediction are orders of magnitude more accurate.

2. The price of AI usage will fall by a factor of 10 every 12 months, and as the price of usage falls, total usage will increase substantially. In fact, between the launch of GPT-4 in early 2023 and the launch of GPT-4o in mid-2024, the price of usage per token will fall by a factor of approximately 150.

3: The social and economic value of AI, which increases performance linearly, will increase super-exponentially. Investment in AI is increasing exponentially, and there is no reason to stop investing.

Altman points out that if these three findings continue, the impact on society will be significant. He also argues that AGI will be the next revolutionary invention for humanity, following the invention of electricity, transistors, computers, and the Internet.

According to Altman, AI will eventually become like a virtual colleague. For example, in the field of software development, AI can perform tasks that would take a software engineer with several years of experience at a top company two or three days to complete, although it still requires human supervision and direction. Furthermore, AI can make hundreds or tens of thousands of people work at the same time. This means that in 10 years, anyone on the planet will be able to achieve more than the 'most influential person right now.'

As AI capabilities improve, the information and energy costs of many products will fall, dramatically lowering the prices of many products, while the value of luxury goods and scarce resources like land could rise dramatically, Altman said.

However, the improvement of AI capabilities is not all good, and there is a risk that the balance between capital and labor will be disrupted. For this reason, Altman argues that measures such as giving everyone on the planet a 'computing budget' to use AI and keeping the cost of using AI as low as possible are necessary.

Altman supports the scaling law that the more the 'computing cost spent on training,' 'the amount of learning data used for training,' and 'the computing cost spent on inference' increase, the more the AI's capabilities improve. However, there are also criticisms that 'the data used for training AI will soon run out, ' and that ' the situation where AI performance improves with increasing training data and computing costs has already ended .' In fact, the ' DeepSeek-R1 ,' released by Chinese company DeepSeek in January 2025, became a hot topic because it had the same performance as OpenAI o1, despite being developed at an overwhelmingly lower cost.

How did DeepSeek beat OpenAI's O1 at 3% of the cost? - GIGAZINE

Immediately after the release of DeepSeek R1, Altman said, 'We believe that computing is more important than ever to the success of our mission. The world wants to use AI a lot, and when the next generation model comes out, it will be truly amazing,' emphasizing his stance that it will continue to be important to spend money on computing. This blog post once again shows the importance of scaling laws.

OpenAI CEO Sam Altman says, 'DeepSeek-R1 is great, but we're going to release even more amazing AI models' - GIGAZINE