OpenAI announced on February 13th local time that it is currently testing the "memory" function of ChatGPT on a small scale: remembering the things discussed by users during conversations and avoiding duplicate information.
According to OpenAI, when chatting with ChatGPT, users can ask it to remember specific content or let it obtain detailed information on its own. The more you use it, the better the memory of ChatGPT will be, and over time, users will begin to notice how "long memory" ChatGPT can bring about improvements.
OpenAI claims that when chatting with ChatGPT, you can ask it to remember some specific things or let it choose the details on its own. The memory of ChatGPT will become better and better, and you will start to notice over time. For example:
You mentioned that you have a toddler who loves jellyfish. When you ask ChatGPT to help her make a birthday card, it will suggest giving you a jellyfish toy wearing a party hat.
You are a kindergarten teacher with 25 students in your class. You would like to schedule a 50 minute course and follow-up activities. ChatGPT will remember this when helping you create a course plan.
OpenAI claims that users can turn off new features at any time. As long as they find the "Settings" menu in the chat GPT user's login interface, click on the "Personalization" option, and set "Memory" to off, ChatGPT will not be able to "remember" the conversation content or use the memorized content.
If users want ChatGPT to forget certain content, just tell it. Users can also find "Manage Memory" in the Personalization option in the Settings menu, where they can view and delete specific memory content, or clear all memories. The memory of ChatGPT changes with user interaction and is not related to specific conversations. Deleting chat records does not erase their memory, it is necessary to delete the memory itself.
OpenAI states that if users want to engage in conversations without using memory functions, they can choose Temporary Chat. Temporary chats will not appear in history, will not use memory functions, and will not be used to train OpenAI models.
For enterprise users, just like other ChatGPT features, users can also control the data of their institution, and information such as memory content in their workplace will not be used for OpenAI model training. Users can control the usage and timing of memory in conversations. In addition, enterprise account owners can close their organization's memory at any time. In the future, enterprise users will have the right to access and retrieve memory content.