AI learns on its own! Is this the beginning of the singularity?

The concept of technological singularity, the point at which artificial intelligence begins to rapidly improve itself, is becoming a reality. Buy AI course. The new ASI-ARCH system is the first in history to complete the entire scientific cycle without human intervention. Not only does it design new AI models, but it also programs them, tests them, and creates even more advanced versions based on the results. We are witnessing the birth of a new evolution, where AI develops completely autonomously, and we are merely observers.

From Theory to Reality: What is Singularity?

Imagine artificial intelligence not as a student who merely performs tasks within given rules, but as a student who starts designing their own subjects and establishing their own schools for even smarter students. This is precisely the principle of technological singularity. Previously, we taught AI, as in the case of chess. Today, this principle is expanding to all fields, and AI is starting to teach us how to “play” everything.

ASI-ARCH: Autonomous Laboratory in Practice

Scientists introduced the ASI-ARCH system, which is the first to complete the entire scientific cycle entirely on its own. It is an autonomous team where AI designs architectures for new models, programs them, tests them, and learns from the results for subsequent generations. Some of these “children” have already surpassed the best human-designed models in real-world tests. Thus, to improve AI, only AI itself is needed.

The system functions as an orchestra of four main roles:

• Researcher: The creative brain that generates new ideas for architectures.
• Engineer: Translates designs into code, tests them, corrects errors, and seeks solutions.
• Analyst: Monitors experiment results, identifies trends, and helps accelerate research.
• Cognitive Base: A vast library of knowledge from both human and AI experiments, drawing, for example, from Hugging Face and ArXiv.

Difference from NAS

Previous methods like Neural Architecture Search (NAS) allowed AI to find the best solutions, but only within the boundaries set by humans. It was like finding a path in a pre-built maze. However, ASI-ARCH breaks these rules. It doesn’t just find a path, but builds new, more efficient mazes itself – designing structures and combinations that a human wouldn’t think of. Within a few weeks, the system conducted over 1,700 experiments and discovered more than 100 entirely new, cutting-edge AI architectures.

Law of Scalability: Progress on Demand

For the first time in history, the so-called “scaling law of discoveries” has been empirically confirmed. Data clearly show that the number of groundbreaking architectures discovered grows directly proportionally to the computational power provided. Double the GPU time means roughly double the new cutting-edge models. Progress is no longer tied to genius or chance, but becomes an industrial process that can be planned by increasing computational resources.

Practical Example: Pathgate Fusion Net

One of the discoveries of the ASI-ARCH system is the Pathgate Fusion Net architecture. It functions as an intelligent dispatcher that evaluates the importance of information in real-time and decides how deeply to send it into the system. As a result, it can deliver the same performance with significantly lower computational power requirements. This efficient mechanism was not conceived by a human, but emerged from autonomous AI experimentation.

Beginning of a New Era

Every improvement ASI-ARCH discovers can be immediately implemented into its subsequent generations. This constantly accelerates the entire self-improvement process. This “infinite loop” is the practical beginning of a true singularity. The question is no longer if it will come, but when. The future, where AI develops itself, begins right now.

What will be our role in this new era of artificial intelligence?