Deep learning works, and it’s expected to get better at scale.
By Sam Altman
Compiled by: Academic Jun
[Editor's Note] Currently, general artificial intelligence (AGI) is a goal that many large model companies at home and abroad are competing to achieve. It is a hypothetical technology that can be comparable to human intelligence in performing many tasks without special training. In contrast, superintelligence goes beyond general artificial intelligence. It can be regarded as a hypothetical level of machine intelligence that can significantly surpass humans in any intellectual task.
Early this morning, OpenAI CEO Sam Altman outlined his vision for future technological progress and global prosperity driven by artificial intelligence in a personal blog post titled "The Intelligence Age."
The article paints a picture of artificial intelligence (AI) accelerating human progress, and he argues that "super intelligence" could emerge within the next few thousand days.
“It’s possible that we could have ASI in a few thousand days (!); it may take longer, but I’m confident we’ll get there,” he wrote.
He believes that AI models will soon become autonomous personal assistants that perform specific tasks on our behalf, such as coordinating medical care for you. At some point in the future, AI systems will become so good that they will help us build better next-generation systems and make overall advances in science.
He attributes what he describes as human society’s next leap in prosperity to deep learning, writing, “Deep learning works, and predictably works better as it scales, and we devote more and more resources to it.”
Altman argues that if we want to get AI into the hands of as many people as possible, we have to make computing cheaper and more plentiful (which requires a lot of energy and chips). If we don’t build enough infrastructure, AI will become a very limited resource that people will fight over, and AI will become primarily a tool for the rich.
Of course, no one can really predict the future of AI, and despite being criticized, as CEO of OpenAI, Altman has probably seen advances in AI that are not yet widely known to the public, so even if he gives a fairly broad timeline, the statement is of concern.
Not everyone shares Altman’s optimism and enthusiasm. “I am very tired of all the AI hype: it has no basis in reality and is simply designed to inflate valuations, stir up the public, grab headlines, and distract from the real work going on in computing,” computer scientist Grady Booch wrote in X.
“Humans are born with a desire to create and use each other, and artificial intelligence will allow us to amplify our abilities like never before,” Altman wrote in the article. “Many of the tasks we do today would have seemed like a waste of time to people hundreds of years ago, but no one will look back and wish they had been a lamplighter.”
Academic Headlines has made a simple translation without changing the main meaning of the original text.
The content is as follows:
Smart Era
In the next few decades, we will be able to do things that seemed like magic to our ancestors.
This phenomenon is not new, but it is accelerating. Over time, people’s capabilities have increased dramatically; we can now do things that our ancestors thought were impossible.
We are more capable not because of genetic changes, but because we benefit from a social infrastructure that is smarter and more capable than any of us individually; in an important sense, society itself is a higher intelligence. Our ancestors—and those before them—created and accomplished great things. They contributed to human progress, and we all benefit from it. AI will give humans the tools to solve difficult problems, helping us add new pillars to the scaffolding that we could not solve on our own. The story of progress will continue, and our descendants will be able to do things that we cannot do.
It won’t happen overnight, but we’ll soon be able to work with AI that can help us accomplish more than we could without it; eventually, each of us will have a personal AI team of virtual experts in different fields, working together to create almost anything we can imagine. Our children will have virtual tutors who can provide personalized instruction in any subject, in any language, at any speed. We can imagine similar ideas improving healthcare, creating any software one can imagine, and more.
With these new capabilities, we can achieve a level of shared prosperity that seems unimaginable today; a future in which everyone will be better off than they are today. Prosperity itself won’t necessarily make people happy—there are plenty of miserable rich people—but it will improve the lives of people around the world.
We can look at human history in a narrow sense: after thousands of years of scientific discovery and technological progress, we have learned how to melt sand, add some impurities, arrange it into computer chips with amazing precision and on an ultra-small scale, run energy through it, and ultimately form systems that can create increasingly powerful artificial intelligence.
This is probably the most important fact in all of history so far. We may have superintelligence in a few thousand days (!); it may take longer, but I believe we will get there.
How do we achieve the next leap in prosperity?
Three words: Deep learning works.
To sum it up in 15 words: Deep learning works, and as the scale increases, the results are expected to get better, and we are investing more and more resources in it.
That's it; humanity has discovered an algorithm that can truly learn any data distribution (or, rather, the underlying "rules" that produce any data distribution) to an astonishing degree of accuracy, and the more compute and data available, the more it can help people solve difficult problems. I find that no matter how much time I spend thinking about this, I can never really understand how important this is.
We still have a lot of details to work out, but it would be a mistake to get hung up on any particular challenge. Deep learning works, and we’ll figure out the rest. There’s a lot we can say about what might come next, but the main point is that AI will get better as it scales, and that will bring meaningful improvements to the lives of people around the world.
AI models will soon become autonomous personal assistants that perform specific tasks on our behalf, such as coordinating medical care for you. Someday in the future, AI systems will become so good that they will help us build better next-generation systems and make overall advances in science.
Technology took us from the Stone Age to the Agricultural Age and then to the Industrial Age. From here, the road to the Intelligent Age is paved with computing, energy, and human will.
If we want to get AI into the hands of as many people as possible, we have to make computing cheap and abundant (which requires a lot of energy and chips). If we don’t build enough infrastructure, AI will become a very limited resource that people will fight over, and AI will become primarily a tool for the rich.
We need to act wisely and firmly. The advent of the intelligent era is a major development, facing very complex and extremely severe challenges. It will not be a completely positive story, but its positive significance is so great that we have a responsibility to ourselves and the future to come up with ways to deal with the risks before us.
I believe that the future will be so bright that no one can describe it now; a distinctive feature of the intelligent era will be massive prosperity.
Although it will all happen gradually, astonishing victories—restoring the climate, establishing space colonies, discovering all of physics—will eventually become commonplace. With nearly unlimited intelligence and abundant energy—the ability to come up with great ideas and the ability to realize them—we can do a lot.
As we’ve seen with other technologies, AI will have downsides, and we need to start working now to maximize the benefits of AI while minimizing its harms. For example, we expect this technology to bring major changes to the labor market in the coming years (both good and bad), but most jobs will change more slowly than most people think, and I’m not worried that we’ll run out of jobs (even if they don’t feel like “real jobs” today). People have an innate desire to create and exploit each other, and AI will allow us to amplify our abilities like never before. As a society, we’ll return to an expanding world where we can once again focus on playing positive-sum games.
Many of the things we do today would have seemed like a waste of time to people hundreds of years ago, but no one would look back and wish they were a lamplighter. If a lamplighter could see the world today, he would find the prosperity around him unimaginable. If we could fast forward a hundred years from today, the prosperity around us would also seem unimaginable.