Artificial intelligence is already beating humans in math olympiads. What does this mean for us?

Artificial intelligence has reached a breakthrough point that seemed like science fiction just a year ago. New models from Google and OpenAI have managed to win the world’s most prestigious mathematical competition, demonstrating the ability to understand and solve complex problems at the level of human elite. This success is not based on mere repetition of learned patterns, but on a new form of learning where AI itself finds ways to solve problems and learns from its own mistakes. This principle, known as reinforcement learning, is already manifesting in programming, where AI can create entire applications based on simple verbal input. Although this raises concerns about the future of professions such as programmers or analysts, history shows that technological revolutions tend to transform work and create new opportunities rather than destroy it.

Live Revolution: AI at the Peaks of Mathematics

Until recently, it was believed that the ability to solve truly complex and creative problems would long remain the domain of humans. However, recent success at the International Mathematical Olympiad refutes this assumption. Two independent artificial intelligence models – Gemini DeepMind from Google and an experimental model from OpenAI – achieved a gold medal by scoring 35 out of 42 possible points. For the first time in history, AI was able to read a problem statement in natural language, understand it, propose solution strategies, and defend the result before human experts, all without any assistance or prior conversion to formal notation.

This leap forward shows that AI no longer merely combines learned examples, but can reason, break down problems into parts, and come up with original and creative solutions. This shifts it from mere imitation to real competition with human intelligence in its strongest disciplines.

How Does AI Learn to Think? Learning from Its Own Mistakes Is Key

What stands behind this rapid progress? The answer is a change in the approach to learning. While older models required precise instructions, the new generation of AI learns autonomously. Thanks to a method known as reinforcement learning, the model can internally generate thousands of possible solutions, test them, evaluate their success, and learn from its own mistakes. Instead of blindly following instructions, AI thus creates its own strategies and procedures.

It’s a process similar to the emergence of an inner voice that evaluates what worked and what didn’t, and tries new combinations. The result is artificial intelligence that finds its own ways and can create solutions that a human alone might not conceive. This approach is beginning to approximate what we call general intelligence – the ability to solve unknown tasks with creativity and adaptation. However, this training is extremely demanding in terms of computational power. For example, a single human-level test, the so-called ARC AGI test, reportedly cost around 300,000 dollars according to some sources.

The Future of Programming: From Code to Intuitive Dictation

One of the areas where these capabilities are most dramatically manifesting is programming. AI today already competes at events like AtCoder World Finals, where it crushes the world’s top programmers. A concept called VIBE coding is emerging, where the user merely describes their idea verbally (e.g., “create a 3D space strategy game”) and AI takes care of the complete implementation – from mechanics design to 3D models and the code itself.

Although today’s results are often still imperfect and full of errors, development is incredibly fast. Just as digital cameras made photography accessible to the masses, AI will make software, application, and automation creation accessible to billions of people without programming knowledge. Writing code is no longer an exclusive club for experts.

Will Artificial Intelligence Replace Us? Change Is Certain, the End of Work Is Not

With the advent of such capable AI, the question of what will happen to human labor logically arises. Many technical and routine professions, such as programmers, analysts, administrators, but also diagnostic doctors or lawyers, will undergo dramatic transformation or even cease to exist. However, history shows us that every technological revolution, from the steam engine to the internet, ultimately created more job opportunities than it destroyed.

The future will not belong to those who can write code, but to those who can effectively utilize new AI tools. Human value will shift from technical execution to rapid adaptation, creativity, and the ability to give meaning to tasks. Agility, curiosity, and continuous learning of new things will be key to success.

The last obstacle on the path to general superintelligence is AI’s ability to understand and evaluate tasks where there is no single correct outcome – for example, in areas driven by feelings, aesthetics, or intuition. Once we can translate these “soft” values into a language that AI understands, entirely new possibilities will open up for it.

What things do you think are currently only evaluable by feeling? And how would you describe that feeling so that AI could understand it and use it as a criterion?