The big language model is not a new product that appeared last year, it can be traced back to the 1990s. GPT-1 was born in 2015. At that time, it did not show a particularly strong ability. Many technology companies were also wrestling in the field of AI. It is also unknown. From the perspective of the time, no one can assert that GPT is the right direction, but in terms of the intelligence emerging from GPT-3, it is indeed much stronger than all previous AI intelligence.
The success of ChatGPT can be seen as an accident.
There are several representatives in the Chinese circle: Wu Jun, Lu Qi, Wu Enda and Li Kaifu. Their different attitudes towards ChatGPT also represent their understanding of this matter. Some people think that ChatGPT is nothing special, and it is a technology that has been proposed long ago. Some people think that it is a new revolution to reconstruct the world, and they will end it in person. No matter what they are engaged in, doing what they can do is beneficial to the industry. promote.
different direction
Both are models, but not always in the same direction, take a look at ChatGLM and WebLLM.
ChatGLM can be privatized and deployed locally. Those companies that are worried about data leakage can privatize the deployment and do an upgrade and development on it, so that security issues and sensitive issues can be guaranteed to a certain extent.
WebLLM is another direction. It allows computing on the client side, relying on the computing power of the Web GPU on the client side, so that small and medium-sized enterprises and individual developers without huge GPU resources can easily explore classes without the cost of computing power. Features of the ChatGPT large language model.
Some companies build on the basis of ChatGPT and do secondary fine-tuning development to train their own small models to meet the needs of the company.
The development of these directions is very similar to the field of cloud computing, where public cloud, hybrid cloud, and private cloud are all applied, and the big language model is the same.