Data Security Classification and Application of GPT
Article directory
foreword
After two or three months of getting familiar with chatGPT, I have some superficial experience in using chatGPT, so let’s reach a consensus first.
- ChatGPT is model trainable and it can create multiple models
- During the conversation, ChatGPT is based on the data set and the context of the conversation.
- ChatGPT's Chinese ability is not very good, the reason can be seen below
- ChatGPT's model training is best based on a specific point of detail. If it revolves around a large frame, the content it provides will also be ambiguous.
1. ChatGPT model
Let’s talk about the first point first. I believe everyone who has used it knows that ChatGPT can build multiple models for special training. There is nothing to say about it.
2. Context association
He will conduct training based on the context of your conversations. When answering for the first time, he will integrate based on the relevant content in the data set. After multiple conversations, he will gradually train into a unique model. (Of course, OpenAI is not free. They keep optimizing some tasteless options, so that in many cases, chatGPT still insists on its own data. If you want to change some of his fixed ideas, you need to try your best to order him )
3. Chinese ability is not very good
Of course, it is a language module made abroad, and there are naturally differences in the use of some Chinese. Many times, it will accidentally write a word wrong, so it is not perfect to rely on it for Chinese work. things
4. How to train a model that suits your needs?
The first paragraph is very important. It lays the foundation for you to choose a data set. When you ask it about a certain knowledge of python, it will find this part of relevant knowledge in the data set. If you continue to ask for details, then it will become an expert in python. But if you ask it instead after asking, do you know Cai Xukun? Its data set will increase, so that at some point, even if the content you ask is very careful, it will have obvious errors.
For example, if I want some business details of securities, I will first let it list the business data related to securities. When it finishes running for the first time, I will let it perform other operations based on the business data, such as the above Each piece of business data is used to describe its usage.
After I have built the framework I need, for example, I need it to list the business data, data types, usage, Chinese pinyin, and English translation of securities, and I am ordering it to generate more relevant business data. Therefore, It will gradually reveal the needs I put forward.
What needs to be remembered is that ChatGPT seems to have a limit on the number of words to reply. When it answers half of it, you can say this: start somewhere and continue to answer. It will be coherent with the above and continue to answer the content in this way.
Summarize
This is just my personal experience, but it does help me gain a lot of convenience in my work. At least I can train some models that I am comfortable with. When I have unknown problems in the same field, I will use ChatGPT in the same field. model to answer some of my questions.