OpenAI is testing the ability for ChatGPT to remember things discussed in conversations to enhance the helpfulness of future chats, placing users in control of ChatGPT’s memory capabilities. OpenAI is experimenting with memory features in ChatGPT, allowing the system to retain information across all chats, which prevents the need for users to repeat information and contributes to more useful future interactions. Users have the authority to manage ChatGPT’s memory, with the option to instruct it to remember or forget specific details either conversationally or through settings, and even to disable the memory feature entirely.
OpenAI is initiating a rollout of these features to a select group of ChatGPT free and Plus users this week to evaluate its effectiveness, with plans to expand access in the near future. The mechanism of memory in ChatGPT involves either direct user instructions to remember specific details or the system’s autonomous information capture, enhancing its memory the more it is utilized. This results in noticeable improvements over time.
For instance, if a user specifies a preference for meeting notes formatting, owns a neighborhood coffee shop, has a toddler with a fondness for jellyfish, or is a kindergarten teacher with specific lesson plan needs, ChatGPT will remember these details to tailor its responses and suggestions accordingly.
Disabling ChatGPT Memory Feature
Users have the capability to disable the memory feature at any point through the settings menu, ensuring that while memory is deactivated, no new memories will be created or utilized. Should users wish for ChatGPT to forget something, they simply need to inform it, and they have the ability to view, delete individual memories, or clear all memories in the settings. It’s important to note that ChatGPT’s memories evolve with user interactions and are not tied to specific conversations; thus, deleting a chat does not erase its associated memories.
OpenAI may utilize content provided to ChatGPT, including memories, to enhance model performance for all users, with options available for users to opt-out through Data Controls. Content from ChatGPT Team and Enterprise customers is excluded from this training process. For conversations where memory usage is not desired, the temporary chat feature serves as a solution, leaving no history, not utilizing memory, and excluding such interactions from model training.
Furthermore, Custom Instructions augment ChatGPT’s utility by allowing users to supply specific details or directives, enabling the system to remember and apply this information in future interactions. With the advent of memory, OpenAI is evolving its privacy and safety standards to address new considerations, such as the types of information that should be remembered and its application, actively working to mitigate biases and prevent the unsolicited memory of sensitive information unless explicitly requested by the user.
Team and Enterprises customers can work more efficiently
For Enterprise and Team users, the memory feature enhances efficiency in work-related uses of ChatGPT, learning user preferences and building on past interactions to save time and deliver more relevant responses. This includes remembering user preferences for tone, voice, format in writing, coding languages and frameworks, and creating customized charts for business reviews, with Enterprise account owners having the ability to disable memory for their organizations, ensuring control over data.
Finally, GPTs will also possess their own distinct memory capabilities, with builders given the option to enable memory for their GPTs, ensuring memories are not shared with builders and requiring users to also enable memory to interact with a memory-equipped GPT. This feature aims to personalize interactions further, such as book recommendations or creating greeting cards, based on remembered preferences, with a broader rollout planned for the future.
Read related articles: