Researchers at Google DeepMind emphasize the need for AI to undergo experiential learning rather than relying solely on limited and static training data. They argue that enabling AI to interact with its environment can enhance its capabilities significantly. Scholars David Silver and Richard Sutton propose a new approach called "streams," which builds upon reinforcement learning, addressing the limitations of current generative AI models like ChatGPT. They highlight the importance of not discarding reinforcement learning to achieve groundbreaking advances in AI technology.
Incredible new capabilities will arise once the full potential of experiential learning is harnessed, say DeepMind scholars, pushing AI beyond the limitations of current data.
The approach suggested by DeepMind, called "streams," aims to remedy the shortcomings of today’s large language models by enabling AI to interact with the world.
Collection
[
|
...
]