UNIT.City — місце, де люди працюють... КРАЩЕ! Обирай свій простір просто зараз 👉
Наталя ХандусенкоAI Eng
23 October 2025, 17:48
2025-10-23
AI agents, supervised by a so-called “parent AI,” have created a new learning system that does not require human intervention.
Like humans, AI learns through trial and error, but traditionally requires human intervention to initiate the process, namely to develop the algorithms and rules that guide the learning. However, as AI technologies develop, machines are increasingly doing this themselves.
An example is a new AI system that invented its own learning method and created an algorithm that outperformed humans on a number of complex tasks.
Like humans, AI learns through trial and error, but traditionally requires human intervention to initiate the process, namely to develop the algorithms and rules that guide the learning. However, as AI technologies develop, machines are increasingly doing this themselves.
An example is a new AI system that invented its own learning method and created an algorithm that outperformed humans on a number of complex tasks.
For decades, human engineers have been developing algorithms that agents use to learn, especially reinforcement learning, where AI learns by receiving rewards for successful actions. While learning is a natural phenomenon for humans and animals, thanks to millions of years of evolution, AI needs to be explicitly trained. This process is often slow and laborious, and is ultimately limited by human intuition.
Inspired by natural evolution, which itself is a process of random trial and error, a group of researchers has created a large digital population of AI agents, writes Tech Xplore.
A special “meta-network” worked on this population — a parent AI that analyzed how successfully agents performed tasks in complex environments and, based on this data, changed the learning rule.
This allowed the system to discover a new learning rule, DiscoRL, which the researchers named Disco57 (because it was evaluated on 57 Atari games), and which turned out to be better than any previously developed by humans.
The team then used Disco57 to train a new AI agent and compared its performance to some of the best human-developed algorithms, such as PPO and MuZero. It was first trained on well-known Atari games, and then on previously unfamiliar tasks, including games like ProcGen, Crafter, and NetHack.
The results were impressive. When running the Atari Benchmark (a set of classic Atari games for evaluating AI), the DiscoRL-trained agent outperformed all human algorithms. When challenged with unfamiliar challenges, it showed the highest level of performance, confirming that the system had discovered its own learning method.
“Our findings suggest that the reinforcement learning algorithms needed for advanced artificial intelligence could soon be automatically discovered based on agents’ experience, rather than being developed manually,” the researchers wrote in their paper published in the journal Nature . “This work takes a step toward creating machine-developed reinforcement learning algorithms that can compete with and even outperform some of the best hand-built algorithms in complex environments.”
Як нейромережі бачать вільну та незалежну Україну? Тест dev.ua
Нейронні мережі для генерації зображень бачать світ по-своєму, їхню логіку зрозуміти часом зовсім неможливо. Але таки хочеться. На честь Дня Незалежності України редакція dev.ua вирішила провести невеликий експеримент.
Ми задали чотирьом різним нейронним мережам п’ять однакових запитів: «прапор України», «День Незалежності України», «український Крим», «перемога України» та «українці». Отриманими результатами ми ділимося з вами нижче.
У TikTok тепер можна генерувати фон за допомогою нейромережі. Ми протестували її та ділимося результатами
У TikTok з’явилася нова функція «Розумний фон». З її допомогою як фон для тіктоків можна підставляти згенеровані нейромережею зображення. Редакція dev.ua протестувала цю технологію і ділиться своїми враженнями.
