In all fairness, as a large-model language chat app, ChatGPT is now very powerful and practical, and can accurately understand the questions raised by users in a very short period of time, and analyze and answer them.
No matter how valuable the answers given by ChatGPT are, its answers can always revolve around the user's questions, and the sentences can be smooth and the ideas are clear, and the overall intelligence is close to that of high school students.
Despite this, there are still many imperfections in ChatGPT at present, for example, there is a father of a seven-year-old daughter, his daughter is about to have a birthday, and the father asks ChatGPT what birthday gift should he give to his daughter? ChatGPT is difficult to answer.
The main reason is not so much ChatGPT's intellectual (technical) problem, but because ChatGPT does not have enough knowledge of the father's daughter's personal situation, which is a very personal, targeted question and there is no uniform standard answer.
To give a satisfactory answer or suggestion, ChatGPT must have a good understanding of the girl's personal situation, such as age, personality, hobbies, including favorite colors, food, ** anime characters and sports, etc., ChatGPT can make a more realistic answer.
When ordinary users ask questions, it is difficult to tell ChatGPT so many details about their daughter at one time, so it is difficult for ChatGPT to come up with a satisfactory answer, and the reason is very easy to understand.
However, the purpose of writing this article is to tell you that at the moment this problem has been preliminarily solved,—— yes, you read that right, this problem is about to be solved, and it is almost "perfectly".
The principle and idea of solving this problem is also very simple, that is, to add a memory function to ChatGPT, and the following is an example of the previous question of "the father asked what birthday gift should be given to his daughter".
Suppose the father has also often communicated with chatgpt before, ** and related topics related to his daughter's learning and growth. In the course of the conversation, the father would sporadically tell ChatGPT some information about his daughter, such as age, personality, hobbies, including favorite colors, food, anime characters and sports, etc.
In the future, ChatGPT will be able to "remember" all the information related to his daughter provided by the father, which is equivalent to building a personal model about his daughter internally.
In this way, when the father asks ChatGPT for advice about his daughter in the future, all the personal information about his daughter does not need to be repeated, and ChatGPT will directly give answers based on the personal model of his daughter established by the previous conversation.
For example, in a previous chat, the father once told ChatGPT that he recently took his daughter to visit the aquarium, and that his daughter was particularly interested in octopuse. Then, when the father asks ChatGPT what his daughter's birthday gift should be, ChatGPT may offer to give his daughter an octopus-style costume, a doll, or a comic book, etc.
For another example, a user once told ChatGPT that he was the owner of a hotel, and the specific location of the hotel, and wanted to do some ** activities during the holidays, and asked ChatGPT for suggestions.
The information that the user has already provided to ChatGPT (such as the specific location of the hotel, etc.) does not need to be provided again, and ChatGPT can directly make suggestions based on the information that has been "remembered" before.
That's the power of adding memory to ChatGPT, which can more intelligently analyze and understand the context and context of the user's questions, so that you can give more pertinent and constructive answers.
Please note that this is not a rumor from any third party, but an official news from OpenAI, on February 13, OpenAI's official website announced that it has officially opened the test to add a memory function to ChatGPT.
Obviously, this feature needs to collect more information provided by users during the chat, which may cause users privacy and security concerns, or even disgust, for which OpenAI will provide several control options for ChatGPT's memory function.
Users can decide and control what ChatGPT remembers and can force him to forget some of the content (see Figures 3 and 4), or simply turn off this feature completely, see Figure 2.
Please note: This feature is currently in beta, and OpenAI says it will roll out to a small group of ChatGPT users starting this week, and then gradually expand its availability.
When it comes to this feature, there is another very important question, that is, after the feature is officially launched in the future, will only ChatGPT paid users be able to enjoy it, and free users can also enjoy it? In this regard, OpenAI has not yet made a clear statement, and will share more relevant revelations as soon as possible, so stay tuned.
In general, the current various artificial intelligence (including ChatGPT) is still in the early stages of development, and there is still huge room for improvement.