We have followed the developments in artificial intelligence for a long time with one question: Is the new model smarter than the old one? In recent months, it has become increasingly clear that this problem alone is not enough. Because the real change is not just that the models respond better. They are simultaneously becoming cheaper, smaller, faster and more independent.
When these four trends are considered together, the real story emerges. Rather than being a separate technology, artificial intelligence can be embedded invisibly into many products and devices, similar to electricity.
INTELLIGENCE IS GETTING CHEAPER
One of the most striking trends today is the rapidly increasing artificial intelligence capacity we get for our money. Models that can do the same level of work operate at a much lower cost than they did a few years ago. Moreover, not only are prices falling, but models are learning to do the same job with less computing power. The economic consequence of this may be greater than we think.
Today, companies still calculate the use of artificial intelligence as a cost item. Every query, every text, every image has a transaction cost. However, if these costs continue to decrease, the question "Is it expensive to use artificial intelligence?" may no longer be meaningful for many applications after a few years.
In this case, artificial intelligence will no longer be an expensive tool used only by large companies, but may become an ordinary feature in everything from accounting programs to educational software, from automobiles to home systems.
The important point here is this: As artificial intelligence becomes cheaper, its use will not decrease, on the contrary, it may explode. When electricity became cheaper, we not only kept the same lamp on for longer. We have produced hundreds of new devices powered by electricity. It would not be surprising to see a similar effect in artificial intelligence.
ARTIFICIAL INTELLIGENCE IS SHINING
One of the more interesting developments that has emerged recently is the physical shrinkage of models. We are not talking about shrinkage in the sense of "his intelligence is decreasing" here. On the contrary, we learn to fit the same talent into a smaller space.
One striking example of this was the recently announced ability to shrink a model with 27 billion parameters to a file of only 5.9 gigabytes. According to its developer's measurements, the model retains most of the performance of its original version, which takes up a much larger footprint. Today, we run most powerful AI in remote data centers. When we ask a question on our phone, most of the heavy work is done elsewhere, in data centers with large computers. But if models continue to shrink, this pattern may change. One day, our refrigerator, car, television, hearing aid or security camera may have its own artificial intelligence. Moreover, it may not need to be connected to the internet all the time. The device can process what it sees or hears on location without sending it to another company.
This is also important in terms of privacy. Camera images, personal documents or health information can be analyzed without going to the cloud. Artificial intelligence can live not only on computers and phones, but on almost any device with energy and processing power.
FROM A QUESTION-ANSWERING MACHINE TO A WORKING MACHINE
Perhaps the most important change is that artificial intelligence is turning from a responsive system into a functioning system. Our usage pattern today is familiar: We ask a question, we get an answer. The goal of new generation systems is different. You tell the result, the system plans the necessary steps on its own. For example, imagine you say, “Research our competitors' activities over the last three months, compare their prices, find any significant changes, and prepare a short report for me.” The new generation system can scan the internet, read the pages, transfer the data to the table, make calculations, check the results and put the report in front of you. Although the difference seems small, it is actually huge. Because here artificial intelligence turns from a consultant to an employee.
This transformation can have serious consequences, especially in white-collar jobs. The lawyer can scan fewer documents and develop more strategies, and the journalist can have artificial intelligence pre-scan them instead of reading hundreds of documents one by one.
Therefore, the real question is “Which jobs will disappear?” It may not be which tasks within each profession will be handed over to machines.
THE SOLUTION, NOT THE MODEL, BECOMES VALUABLE
Today, the technology world is debating which artificial intelligence model is better. OpenAI, Google, Anthropic or Chinese models? But the gap between the most powerful models is narrowing. Openly available models are also much more competitive than they were a few years ago.
If this trend continues, after a few years most users may not even need to know which model works. Just as we do not think about the technologies behind a site today, we may not think about which artificial intelligence model we connect to tomorrow.
In such a world, the real value shifts to places other than the model itself. Who knows the user better? Who integrates better into the business process? Which system is reliable? Which company has better proprietary data, distribution power and customer relationships? For this reason, the winners of the coming period may not necessarily be those who make the greatest artificial intelligence model.
Those who best use AI to solve a real problem can also create great value.
ARTIFICIAL INTELLIGENCE ACCELERATES ITS OWN DEVELOPMENT
Less talked about, but perhaps the most important development in the long term, is this: Artificial intelligence is now turning into a working tool for people who develop artificial intelligence.
Software engineers, researchers and data scientists are making more and more use of artificial intelligence. The result of this may be greater than a simple productivity increase.
Better AI allows researchers to work faster. When researchers work faster, better artificial intelligence is developed in less time. The new system also speeds up the next search even further. We are not talking about a completely self-improving machine here. People are still at the center of the process (at least for now). But a feedback loop has begun to form that increases the pace of development. Therefore, it may be misleading to think of the progress of artificial intelligence in the next five years as a straight extension of the pace of the last five years. The rate of development itself may also vary.
NEW SCORE RESOURCE: TRUST
When artificial intelligence starts to produce too fast, another problem arises: Who will check that what is produced is correct? Today, an artificial intelligence can produce pages of text, hundreds of lines of code or a detailed analysis in a few seconds. Soon scientific hypotheses, legal assessments and complex business plans could also be produced much more quickly.
But when the production rate exceeds the verification speed, the bottleneck changes. The problem used to be finding enough people to prepare a report. The problem in the future may be to figure out which of the hundred reports produced by AI are actually true.
Therefore, one of the important areas of the future may not be “more artificial intelligence” but reliable artificial intelligence. Systems that can show their sources, control their own work, and ask for human approval when necessary will gain value. It would not be surprising if competition in many fields, from journalism to health, is based on trust rather than intelligence.
ARTIFICIAL INTELLIGENCE IS LEAVING THE COMPUTER AND ENTERING THE WORLD
The next big step could be robots. Most robots today are very good at repeating certain movements, but have difficulty when faced with a new environment. Bringing the latest developments in artificial intelligence to robots is trying to change this limit. The aim is to move from robots whose every movement is pre-programmed to robots that can understand what they see and act according to the given task. If this can be done, the use of robots could expand beyond factories. Warehouses, hospitals, hotels, restaurants, farmland and eventually homes could be next in line. The new artificial intelligence model called Helix 2.5, developed by the humanoid robot company named Figure, gives promising signs in this regard. He leaves the robots using Helix 2.5 to homes they have never seen before (that is, they have not bent), and the robots can create wonderful works in these houses.
However, we should expect slower progress from the software here. A chat program giving the wrong answer is not the same as a 70-pound robot making the wrong move. It will not be easy for robots to spread into daily life without solving safety, energy, maintenance and cost problems. But in the long run, one of the biggest economic impacts of artificial intelligence may emerge here. Because when software intelligence is connected to physical labor, not only office work but also millions of tasks performed in the real world begin to change.
THE REAL CHANGE MAY HAVE YET STARTED
What all these developments have in common is this: AI is simultaneously becoming smarter, cheaper, smaller and more independent.
If there was only one of these four trends, its impact would be limited. But when they all proceed together, a different picture emerges.
Today, we see artificial intelligence as a service on our phone or computer. Tomorrow we may not even notice it. While our car plans the journey, our home manages energy, our doctor examines the results, the teacher prepares a special study for the child, and our company carries out daily operations, there may be small artificial intelligences constantly working in the background.
Perhaps, in a few years, the question "Do you use artificial intelligence" will seem as strange to us as the question "Do you use the Internet" today? Because the real transformation is not the introduction of artificial intelligence into our lives. It will become invisible.
