On the other side of the ocean, Anthropic, which released the Claude series of large models, has always been seen as OpenAI's strongest competitor. In the eyes of the outside world, its AGI image is very different from OpenAI: it pays more attention to safety and risks, which is closely related to the research style of Anthropic's founder and CEO, Dario Amodei.

However, in a recent article published by Amodei, he rarely mentioned the "huge potential of AI". In this long article of ten thousand words, Amodei actively predicted that "powerful AI" (similar to AGI) may come as early as 2026. In pure intelligence fields such as biology, AI can even surpass Nobel Prize winners. He believes that powerful AI can increase the speed of scientific discoveries by 10 times, and compress 50-100 years of biological progress into 5-10 years, curing most diseases and extending human life to 150 years.

At this point in time, some market voices speculate that the article may be related to Anthropic's financing and new product hype. At the end of September, foreign media reported that Anthropic has begun discussing new rounds of financing with investors, with a valuation that may reach as high as $40 billion, roughly doubling the valuation after this year's financing. As a major investor in Anthropic, Amazon has invested about $4 billion and may continue to support in the new round of financing.

Advertisement

At the same time, there is also news that Anthropic has disclosed to investors that it expects its operating income for this year to reach $1 billion. This means that Anthropic's revenue will grow by 1100% year-on-year. It is reported that 60% to 75% of Anthropic's revenue comes from third-party application programming interface (API) access, that is, the revenue from a large number of developers and third parties such as Amazon Cloud AWS calling Anthropic models accounts for as much as three-quarters.

"Powerful AI" may come as early as 2026.

"As the CEO of Anthropic, some people may mistakenly think I am a pessimist, but the fact is not so." Amodei said that most people both underestimate the huge potential of AI and underestimate the serious risks that AI may bring.

In the long article, Amodei did not forget to mention Anthropic's consistent theme - risk. In the article, he frankly said that he was disgusted with the attitude of some public figures in the field of AI risks (not to mention leaders of AI companies) towards the post-AGI era, as if their task was to lead humanity to redemption alone. "I think it is very dangerous to regard the company as the only force that can change the world, or to deify technological goals as religious missions."

Amodei said that too much "self-promotion" is not beneficial to an individual's mentality. If AI companies always exaggerate the various benefits of AI, it may make people feel like they are engaging in propaganda or trying to cover up the adverse side. He believes that AI should avoid "science fiction" color.

Amodei also denied the expression of AGI (Artificial General Intelligence), "I find AGI to be an imprecise term, it gathers a lot of science fiction baggage and hype." He prefers to use "powerful AI" or expert-level scientific engineering to express this meaning.

The "powerful AI" that Amodei refers to is also an AI model, which may be similar to today's large language models (LLM), but may be based on different architectures, involve multiple interactive models, and use different training methods. In terms of specific features, Amodei explained that in terms of pure intelligence, powerful AI will be smarter than Nobel Prize winners in fields such as biology, programming, mathematics, and writing. This means that it can solve unsolved mathematical theorems, write excellent novels, and develop complex code libraries from scratch.Focusing on risks does not mean negating the future of AI. Many people are skeptical about the imminent arrival of powerful AI, and some even believe it may never come to pass. However, Amodei thinks it could arrive as early as 2026, but it might also take longer.

Regarding how future AI might change the world, Amodei highlighted its positive applications in areas such as biology and physical health, neuroscience and mental health, economic development, and poverty. "My predictions may seem radical, but I'm serious," said Amodei.

Amodei earned his undergraduate degree in physics from Stanford University and his doctorate in biophysics from Princeton University, before returning to Stanford for three years as a postdoctoral fellow. "I have a professional background in biology and neuroscience, and I also understand some economic development." It is based on such a background that Amodei has more to say about the combination of biology and AI.

In the field of biological science, Amodei believes that AI should not only be seen as a data analysis tool, but as a "virtual biologist" capable of doing all the work that biologists do, including designing and conducting experiments, inventing new biological technologies and measurement methods, etc. AI, by accelerating the entire research process, can truly promote the development of biology.

Based on the above possibilities, Amodei predicts that powerful AI can at least increase the speed of scientific discoveries by 10 times, allowing us to achieve biological progress equivalent to 50 to 100 years within 5 to 10 years. He calls it the "compressed 21st century": that is, within a few years after the emergence of powerful AI, we will achieve progress in the field of biology and medicine that would take the entire 21st century.

Why not faster, such as increasing the speed by 100 times? Amodei explained that sequence dependence and experimental time are important: to achieve 100 years of progress in 1 year, everything needs to be executed perfectly the first time, including animal experiments, microscope design, and the construction of expensive experimental facilities, etc. Experiments and hardware design have a fixed "response time", and a certain "number of necessary iterations" are needed to draw some conclusions.

Human lifespan will be extended to 150 years

If the aforementioned goals are achieved within the next 7-12 years, Amodei said that this would be able to eliminate most of the diseases that have plagued humanity for thousands of years, and human lifespan would double to 150 years. When the current generation of children grows up, they will enjoy more biological freedom and the right to self-expression, and even if they are lucky, perhaps their lifespan can be as long as they want.

Regarding the view that human lifespan will double, Amodei mentioned that it seems radical, but in the 20th century, life expectancy increased nearly twice (from about 40 years to about 75 years), so the "compressed 21st century" will double again to 150 years is a trend.

Specifically, there are already drugs that can extend the maximum lifespan of rats by 25%-50% with limited side effects. And once human lifespan reaches 150 years, it is possible to reach the "escape velocity" (referring to the speed of extending life exceeding the speed of aging), to win enough time for most people alive today.In addition to biology, Amodei believes that the framework of "achieving 100 years of progress in 5-10 years" also applies to neuroscience. In the future, humans are also expected to find treatments or prevention methods for most mental illnesses within 100 years — if accelerated by AI, it may only take 5-10 years, and most mental illnesses such as PTSD, depression, schizophrenia, and addiction may be cured.

At the same time, in terms of economic development, AI-accelerated energy technology, improved building materials, and better logistics systems may naturally spread through market mechanisms, just like the proliferation of mobile phones in sub-Saharan Africa. "An ideal vision is to achieve an average annual GDP growth rate of 20% for developing countries, with 10% coming from AI-driven economic decisions and another 10% from the natural spread of AI technology." If achieved, Amodei said that this would raise the per capita GDP of sub-Saharan Africa to the current level of China within 5-10 years, while pushing other developing countries to an economic level higher than the current United States. Of course, this is an ideal vision that requires some effort to achieve the goal.

In addition, Amodei also mentioned the application of AI in food safety and climate change in this long article. For example, by increasing crop yields and optimizing agricultural supply chains, AI may drive a second green revolution. In addition, AI may be able to help discover technologies for removing carbon dioxide from the atmosphere and clean energy technologies.

Although this is indeed a future worth striving for, the vision mentioned in the article also seems very radical. Even Amodei himself said, "Almost no one would think that such a transformation will be achieved within the next decade, and some may even think it is a fantasy." On the other hand, he believes that civilization has successfully experienced significant economic transformations in the past, from hunting and gathering to agriculture, from agriculture to feudalism, and from feudalism to industrialization. In Amodei's view, AI may be a variable in the future.