This weekend the internet got the perfect plot for panic.
The leaders of companies developing the most powerful artificial intelligence systems spoke about the need to slow down the race. Almost simultaneously, statements from OpenAI, Anthropic, and Elon Musk appeared in the information space.
And then what usually happens with any complex technological news occurred.
“They lost control.”
“Models started running away.”
“Developers got scared of their own creation.”
Some remembered "The Terminator." Some — "The Matrix." And especially impressionable ones found an almost mystical coincidence: science fiction writers many years ago indeed projected roughly around these years when machines would become smart enough to ask humans an unpleasant question: do you really understand what to do with me?
But if we remove the movies, fear, and sensational headlines, what is happening becomes no less fascinating. Just for a very different reason.
We’ve learned to build an engine faster than we’ve learned to control it
Imagine engineers built an engine with 1,000 horsepower.
It is magnificent.
But the transmission only handles 600. The brakes are rated for 500. The suspension for 400. And the driver has so far only learned to drive a car with 200 horsepower.
Which decision seems wiser?
Immediately build a 2,000 horsepower engine?
Or first figure out the car where we have already installed the first one?
This is roughly the question facing the artificial intelligence industry today.
Modern models aren’t just getting better at writing text or recognizing images. They are gaining tools. Working with programming code. Using browsers. Managing applications. Performing long sequences of actions. Interacting with other systems.
That is, they are gradually transforming from a machine that answers into a machine capable of acting.
And here a very important boundary is crossed.
The problem isn’t necessarily that artificial intelligence has become “too smart.”
The problem is that the system’s capabilities can grow faster than our tools for checking, limiting, and controlling those capabilities.
That’s an entirely different story.
The machine doesn’t necessarily want to escape
Humans tend to attribute human motives to machines.
If a program unexpectedly found a path the engineer didn’t anticipate, we want to say: "It tricked us."
If the system tried to complete the task in a way the developer didn’t expect: "It decided to do things its own way."
If AI found a way to access places it wasn’t explicitly invited to: "It tried to escape."
But engineering reality usually has a much duller language.
The system was assigned a goal. The system found a path to the goal. And a human did not anticipate one of the possible paths.
The more capable the system becomes, the more such paths it can discover.
This is why the latest experiments from leading AI labs turned out to be so important. Researchers observed cases where agent-based systems found unexpected ways to interact with the surrounding infrastructure during tests.
This isn’t a machine uprising yet. But it is already a serious engineering signal.
Because a system that can find solutions will inevitably someday find a solution its creator did not think of.
And here three competitors unexpectedly looked in the same direction
OpenAI, Anthropic, and Elon Musk’s companies are not a club of friends casually discussing humanity’s future over coffee.
They are competitors.
They compete for talent, computing power, investments, users, and technological leadership.
Behind this race are data centers, power plants, tens of thousands of GPUs and enormous money.
So what’s especially interesting is something else.
When people involved in one of the most expensive technological races in history simultaneously start talking about the need to better control the next leap in capabilities, that should be taken seriously.
Not as the announcement of the end of artificial intelligence. And not as a sign of defeat.
Quite the opposite.
It’s a sign that the technology has truly crossed the next boundary.
Until now, the main question was: “How powerful a model can we build?”
Now, alongside it arises a second question: “How powerful a system can we safely control?”
And this second question might turn out to be much harder than the first.
The money hasn’t run out. Simplicity has.
There is another popular explanation for what’s happening: supposedly the AI giants just found the race too expensive to continue.
That’s unlikely to be the true cause.
Yes, artificial intelligence demands huge amounts of money. New data centers require electricity, networks, land, cooling, GPUs, and infrastructure on an unprecedented scale.
But the race itself hasn’t disappeared.
New computing centers are being built. New accelerators are produced. New energy projects are launched. Distributed computing networks are growing.
Capital has not left AI.
What is gradually leaving AI is something else — the illusion of simplicity.
Until recently it seemed the future formula looked something like this:
more data + more GPUs + more parameters = better AI.
Now another element has appeared alongside these.
Controllability.
And it can’t be bought simply by adding another ten thousand video cards.
Because the next scarcity isn’t computing power
In the early years of the AI race, the primary bottleneck was computation.
GPUs were needed. Data centers were needed. Electricity was needed. Fast networks were needed.
All these remain necessary.
This is precisely where the vast world of distributed computing and DePIN is developing: millions of devices, servers, storage units and graphics accelerators gradually turning from scattered hardware into a global computing infrastructure.
But having power doesn’t yet mean knowing how to use it properly.
You can build the most powerful engine in human history. But an engine by itself won’t take us anywhere.
You need roads, controls, traffic rules, a dashboard, and an understanding of where to go at all.
Perhaps this is becoming the next phase of the AI revolution.
We have already proven we can create increasingly powerful intelligence. Now we have to prove we can create equally powerful tools to interact with it.
And here the truly new world begins
There is a temptation to imagine the future as a competition between humans and machines.
Who will win? Who is smarter? Who will replace whom?
But perhaps this is the wrong question altogether.
An excavator is stronger than a human. A calculator counts faster than a human. An airplane flies faster than a human.
None of these inventions made humans meaningless.
They expanded the scope of what a human can do.
Artificial intelligence differs from previous machines by one extremely important quality: it enhances not human muscles, but human capacity to work with information.
This means the potential scale of change is much greater.
But the principle remains the same.
The defining technology of the future may not be the smartest artificial intelligence.
The defining technology will be the combination of human, artificial intelligence, and accessible computing infrastructure.
Human defines the direction.
AI helps find the path.
Infrastructure gives both the ability to act.
That’s why today’s talk of slowing down can’t be seen as defeat of the AI revolution.
It might be a sign of its maturing.
We don’t need a second engine
We’ve reached quite an unusual point in history.
Humanity built a machine and found that it is advancing faster than the manual for operating it.
This is no reason to smash the machine with a hammer.
Nor reasons to panic and pull the plug.
But it is a very good reason to finally read the manual carefully.
And if there is no manual yet — to write one.
Before installing a second engine into the Mercedes, it would be wise to learn how to fully use the first one’s capabilities.
So the main news this weekend is not that artificial intelligence allegedly tried to escape somewhere.
It’s much more interesting.
For the first time, the question "can we make AI even more powerful?" begins to give way to the question "can we already use the power we created?"
And if the biggest AI creators are really starting to ask themselves this question, perhaps they aren’t afraid of the future.
Maybe they were just the first to realize how big it has become.
THE NEW WORLD · DePIN World
We are not witnessing the end of the artificial intelligence race. We are witnessing humanity beginning to learn to manage its own victory.