foreword
This article mainly talks about some introductions about ChatGPT in general.
First of all, ChatGPT and GPT are not the same thing, don't confuse them. The former is a product and the latter is a language model. ChatGPT is a product developed on the basis of GPT.
The most powerful ChatGPT is a conversational product developed based on the super-large data model of GPT. It has reached GPT3.5, and it is not as magical as it is said on the Internet. Bloggers' eyeballs earn traffic.
Careful students will find that there are still many problems, such as the inability to directly use voice dialogue (if you want to receive voice, you have to open it), there are too many limitations due to the use of the scene, and there are big problems with the accuracy of the content (such as The returned image address is often 404) and so on.
Many people began to slowly mention whether AI will replace human beings in the future. I suggest that you just smile when you see such words. According to the current development of science and technology, at least we will not encounter this situation in our lifetime. As for thousands of years later, we don't care if we want to worry about it. We live in the present and enjoy ourselves in time.
1）AI 本身是把双刃剑。给人类带来一定的便捷的同时，也会让一些人下岗失业，这是人类社会文明必然会出现的现象。；
2）但别惊慌，更别排斥它，AI 归根到底还是科技与狠活儿，套用行业大佬的一句话就是：无能的人才会去惧怕它，而聪明的人会学会利用 AI 。计算机的发明让很多人失业了，但是造福了更多的人。AI 也是如此！！！
3）AI 的出现只会让我们 ”终身学习” 的信念更加坚定。（人如果不学习，那跟猪有什么区别？人生还有何意义呢？）
1. What is ChatGPT
ChatGPT is a chat robot technology developed by OpenAI based on a language model.
（应用场景：翻译、问答、代码生成、文本摘要、推文分类等）
2. Basic knowledge of GPT
The full name is: Generative Pre-Training (generative training)
其发展历程（收集—>汇总训练—>引导训练）
1）GPT-1
- Pre-Training, pre-training
- Fine-Tuning, supervised model (ordered)
2）GPT-2
- (Expand training scale: WebText dataset, 1.5 billion parameters)
- unsupervised model
3）GPT-3
- Larger training model, 175 billion parameters (Sparse Transformer, alternately dense and sparse)
- Bootstrap training, training on a small number of samples
3. Basic knowledge of NLP
1) Concepts and techniques of natural language processing
- Natural Language Understanding (NLU)
- Natural Language Generation (NLG)
2) Main content
- text mining
- information retrieval
- Syntactic Semantic Analysis
- machine translation
- question answering system
- dialogue system
4. GPT-3 model architecture
- text-ada-001 (fastest, cheapest, only very simple tasks (2048))
- text-babbage-001 (very fast, less expensive, only for simple tasks)
- text-curie-001 (faster than davinci, less costly, very capable)
- text-davinci-003 (higher quality, longer output, stronger command understanding, and support for text insertion and completion (4000))
5. Related documents
- Official Documentation (Science Online) (https://platform.openai.com/docs/introduction)
- Account registration process (https://www.cnblogs.com/chat-gpt/p/how-to-use-chatgpt-in-china.html)
- Idea joins ChatGPT (https://docs.obiscr.com/article/GPT-Getting-Started)
- View api-key (Science Online) (https://platform.openai.com/account/api-keys)
- WeChat access to ChatGPT (https://51.ruyo.net/18261.html)
- Download the installation package for accessing WeChat (https://gitee.com/shtml/wechatbot/tree/main/bin)