Hello, today is a very important news from the IT world. Today I will tell you about a new revolutionary AI, which will radically change our lives in the future. The question is which way?
Not so long ago, we published a post about GPT-3, a revolutionary neural network that learned to "understand" the meaning of a text thanks to training with huge amounts of data. Why is GPT - 3 so important? The question is obvious, answer it yourself. So far, ordinary people are actively blurring their eyes with deepfakes, writing essays and generating fake cats. I assure you, this is the least that these neural networks of the future can do. Imagine an AI that will write any software for you, having learned only a few key parameters from the technical task. At this moment, give free rein to your imagination, let it play out pretty well.
But, not the point. Just the other day, GPT-3 was dropped by its Chinese counterpart. Scientists from the Beijing Academy of Artificial Intelligence rolled out the WU Dao 2.0 neural network, which is about 10 times larger than GPT-3, having learned 1.75 trillion parameters. Okay, it's an order of magnitude larger, but what's the point. The Chinese have also created the latest training system, which is an order of magnitude more efficient and convenient. And the most important thing. While GPT-3 and a neural network from Google Switch Transformer are sharpened either specifically for text or for face recognition, you can do anything you want with a Chinese monster.
Wu DAO can process natural language, generate text, recognize images and create them. The model can not only write essays, poems and couplets in Chinese, but also generate alternative text based on a static image and almost photorealistic images based on descriptions in natural language. Wu Dao has also demonstrated his ability to predict the three-dimensional structures of proteins such as AlphaFold. Plus, the Chinese neural network needs a lot fewer examples to train. For example, to answer questions about the text read, GPT-3 will need 570 gigs of examples, and Chinese AI will need only 10 to 100 examples.
Not so long ago, we published a post about GPT-3, a revolutionary neural network that learned to "understand" the meaning of a text thanks to training with huge amounts of data. Why is GPT - 3 so important? The question is obvious, answer it yourself. So far, ordinary people are actively blurring their eyes with deepfakes, writing essays and generating fake cats. I assure you, this is the least that these neural networks of the future can do. Imagine an AI that will write any software for you, having learned only a few key parameters from the technical task. At this moment, give free rein to your imagination, let it play out pretty well.
But, not the point. Just the other day, GPT-3 was dropped by its Chinese counterpart. Scientists from the Beijing Academy of Artificial Intelligence rolled out the WU Dao 2.0 neural network, which is about 10 times larger than GPT-3, having learned 1.75 trillion parameters. Okay, it's an order of magnitude larger, but what's the point. The Chinese have also created the latest training system, which is an order of magnitude more efficient and convenient. And the most important thing. While GPT-3 and a neural network from Google Switch Transformer are sharpened either specifically for text or for face recognition, you can do anything you want with a Chinese monster.
Wu DAO can process natural language, generate text, recognize images and create them. The model can not only write essays, poems and couplets in Chinese, but also generate alternative text based on a static image and almost photorealistic images based on descriptions in natural language. Wu Dao has also demonstrated his ability to predict the three-dimensional structures of proteins such as AlphaFold. Plus, the Chinese neural network needs a lot fewer examples to train. For example, to answer questions about the text read, GPT-3 will need 570 gigs of examples, and Chinese AI will need only 10 to 100 examples.
