Recently, ChatGPT has become a sensation on social networks, and many people are eager to experience this new phenomenon. There are already many articles explaining how to register for ChatGPT, so we won’t dwell on that. Here, I will provide a simple overview to help readers quickly understand ChatGPT and the AI technology behind it in just a few minutes.
GPT stands for “Generative Pre-trained Transformer”, which aims to use deep learning to generate human-understandable natural language. Currently, when we refer to GPT, we generally mean GPT-3. It is evident that there were previous versions, GPT-2 and GPT.
GPT-3 was trained and developed by the artificial intelligence company OpenAI. The model is designed based on the transformer language model developed by Google. The neural network of GPT-3 contains 175 billion parameters, making it the largest neural network model ever created. OpenAI published the paper on GPT-3 in May 2020, and Microsoft announced on September 22, 2020, that it had obtained exclusive licensing for GPT-3.
According to OpenAI, “We have trained a model called ChatGPT that interacts in a conversational manner. The conversational mode allows ChatGPT to answer follow-up questions, acknowledge mistakes, challenge incorrect premises, and refuse inappropriate requests. ChatGPT is the sibling model of InstructGPT, which was trained to follow instructions in dialogues and provide detailed responses.”
ChatGPT is a model optimized based on GPT-3.5 and can be understood as a general-purpose chatbot. According to OpenAI, GPT-3.5 learns the relationships between sentences, words, and parts of words by absorbing a vast amount of content from the internet, including thousands of Wikipedia entries, social media posts, and news articles.
Readers can unleash their imagination to have ChatGPT complete various creative tasks. For example, asking it to write a summary report in the style of Lu Xun or to create copy in the style of Jin Yong.
Additionally, it is worth noting that Prompt Engineering is currently a hot area of research and may become a dedicated career path in the future. Interested readers can explore this further.
ChatGPT essentially belongs to generative artificial intelligence and falls under unsupervised or semi-supervised machine learning. Related to this is Discriminative modeling, which mostly belongs to supervised learning.
There are currently two main frameworks for generative artificial intelligence: GAN (Generative Adversarial Network) and GPT (Generative Pre-trained Transformer).
GAN is widely used in image, video, and speech generation, with practical applications in fields such as healthcare, autonomous driving, and the metaverse.
With the introduction of GPT-4, generative artificial intelligence is expected to once again exceed people’s expectations.
