Recently, major breakthroughs have been made in the field of artificial intelligence! The Genius agent developed by the Verses team has demonstrated amazing learning capabilities and gaming proficiency in the classic game Pong, surpassing top human players and other AI models with only 10% of the data and 2 hours of training time. This not only sets a new record for AI learning efficiency, but also provides new inspiration for the future research and development direction of AI agents. Its efficient learning mechanism and initiative deserve in-depth study.
Recently, the Genius agent developed by the Verses team has achieved amazing results in the classic game Pong, surpassing top human players and other AI models with only 10% of the data and 2 hours of training time. This breakthrough marks a new milestone in AI technology and heralds the development direction of future intelligent agents.
The success of the Genius agent is inseparable from its unique design concept. Compared with traditional large models, Genius is only 4% the size of the SOTA model IRIS and can run on ordinary M1 chip MacBooks. The researchers were inspired by an experiment four years ago. Scientists found that a cultured "brain on a plate" could learn the Pong game in just 5 minutes, which triggered their thinking about imitating the way the human brain works.
The Verses team believes that traditional large model-based AI agents have serious deficiencies in logical reasoning. Existing models rely more on memorizing inference steps from training data and lack true initiative and curiosity. The Genius agent adopts the concept of cognitive engine, which not only has cognitive, reasoning and decision-making capabilities, but also gives the agent the ability to actively learn.
In comparative tests with IRIS and other AI models, Genius demonstrated strong learning capabilities. The researchers trained Genius using 10,000 steps of game data in 2 hours, and the results showed that its performance exceeded IRIS, which was trained for two days. The success of Genius lies not only in its ability to learn quickly, but also in its active performance in games. For example, in the Pong game, Genius was able to come back and win even after falling behind, a phenomenon that had never occurred in IRIS training.
However, researchers also cautioned that although Genius's performance is exciting, there is currently a lack of unified standards that can comprehensively measure AGI performance, and diversified tests are needed to verify its adaptability and reliability in different fields.
This research result not only promotes the development of AI agents, but also provides new ideas and methods for future machine intelligence exploration.
Paper address: https://arxiv.org/pdf/2410.05229
The success of the Genius agent has opened up a new path for AI research. Its efficient learning mechanism and imitation of the human brain mechanism are worthy of further research and exploration. In the future, similar lightweight and efficient AI models are expected to play a role in more fields and promote the continued progress of artificial intelligence technology.