Current AI capabilities delay the creation of a machine with a human brain.
Google expressed doubts about the implementation of Artificial General Intelligence (AGI), which is often considered the Holy Grail in the field of AI. In a new article, the corporation pointed out the limitations of the transformer technology underlying large language models such as ChatGPT. The main problem is the inability of transformers to generalize information beyond their training data, which calls into question their effectiveness in new, unfamiliar tasks.
The article notes that while transformers can effectively work with tasks that are closely related to their training data, they cannot solve tasks that go beyond their stored data. The discovery is critically important for those who seek AGI-a hypothetical intelligence that can perform any actions like a human.
Google stressed that current advances in AI should be evaluated based on their limitations, especially in light of recent scientific discoveries. According to Google, despite significant investment and interest in AGI, it is necessary to recognize that general artificial intelligence is still far from being implemented.
An OpenAI representative said that to reach the AGI level, artificial intelligence must perform many generalizing tasks that the human brain can perform — whether it's adapting to unfamiliar scenarios, creating analogies, processing new information, or thinking abstractly.
An NVIDIA representative expressed surprise at why the discovery of transformers limitations came as a surprise to many, pointing out the fundamental rule of machine learning about the need to distribute tests during training.
While some may see this as a cause for concern, Lamini AI sees it as an opportunity to develop and improve AI by training models in new environments, rather than just using them in their current state. Therefore, AI can still be extremely useful and effective, despite its limitations.
Google expressed doubts about the implementation of Artificial General Intelligence (AGI), which is often considered the Holy Grail in the field of AI. In a new article, the corporation pointed out the limitations of the transformer technology underlying large language models such as ChatGPT. The main problem is the inability of transformers to generalize information beyond their training data, which calls into question their effectiveness in new, unfamiliar tasks.
The article notes that while transformers can effectively work with tasks that are closely related to their training data, they cannot solve tasks that go beyond their stored data. The discovery is critically important for those who seek AGI-a hypothetical intelligence that can perform any actions like a human.
Google stressed that current advances in AI should be evaluated based on their limitations, especially in light of recent scientific discoveries. According to Google, despite significant investment and interest in AGI, it is necessary to recognize that general artificial intelligence is still far from being implemented.
An OpenAI representative said that to reach the AGI level, artificial intelligence must perform many generalizing tasks that the human brain can perform — whether it's adapting to unfamiliar scenarios, creating analogies, processing new information, or thinking abstractly.
An NVIDIA representative expressed surprise at why the discovery of transformers limitations came as a surprise to many, pointing out the fundamental rule of machine learning about the need to distribute tests during training.
While some may see this as a cause for concern, Lamini AI sees it as an opportunity to develop and improve AI by training models in new environments, rather than just using them in their current state. Therefore, AI can still be extremely useful and effective, despite its limitations.
