Researchers from the University of Tsinghua have developed a new model for Amnesty International that can train themselves without any humanitarian statements. This penetration, called “The absolute zero(AZR) opens new opportunities in artificial intelligence research and can lead to the most independent and capable artificial intelligence systems in the future.
- Self -learning capacityAZR can be developed and improved without human involvement. He learns by performing tasks and verifying their results through a process that is rewarded.
- Do not depend on external dataThis method is unique because it does not require “gold stickers” (pre -specific data) or humanly specific problems for high level.
- Constant improvement: By performing tasks and obtaining notes on their results, AZR can constantly adjust their algorithms and improve them, Which can lead to the development of Superinteigence.
This new method allows artificial intelligence systems to create your exercise data and learn through the continuous counterattack loop.
Without entering human data
Imagine the pianist who learns to play without teachers, notes or recordings – only by trying the keys and listening to the result. Likewise, the absolute zero cause works completely without human data. Traditional artificial intelligence systems are similar to students who need thousands of examples of teachers to learn, but AZR breaks this series of addicts using the “Self -Islam Episode”. The system turns between rotation: one suggests challenges (coding problems, mathematical equations) and the disease that tries to solve them. The Python Run icon works as an objective judge that determines whether the solutions are correct, which gives the model direct notes without human intervention.
Despite promising results, there are still challenges. AZR is in its cradle and researchers suggest that there is room for improvement, especially in the most complex thinking. In addition, self -learning completely raises moral questions about control and supervision.
