ChatGPT Introduces 'Memory' Feature to Recall Your Preferences and Identity

Tech News February 14, 2024

Engaging with an AI chatbot can sometimes feel repetitive, like reliving the same scenarios in "Groundhog Day." You find yourself repeatedly specifying your email preferences and recounting the weekend activities you've already completed multiple times. OpenAI aims to address this by introducing a significant enhancement to its bot's capabilities. The latest feature, called "memory" for ChatGPT, enables the bot to retain information about you and your past conversations.

Memory operates in two ways. You can instruct ChatGPT to remember specific details about you, such as your preference for coding in Javascript, your boss's name (Anna), or your child's sweet potato allergy. Alternatively, ChatGPT can autonomously gather and store information about you as you interact, gradually building a personalized knowledge base. The objective is to make ChatGPT more personalized and intelligent, eliminating the need for repetitive reminders.

Each customized GPT instance comes with its own memory. For instance, using Books GPT as an example, with memory activated, it can automatically recall the books you've read and your preferred genres. The potential applications of this memory feature are vast, ranging from educational platforms like Tutor Me optimizing long-term courses based on your knowledge to travel services like Kayak swiftly accessing your preferred airlines and hotels. Additionally, fitness apps like GymStreak could use memory to track your progress over time.

The introduction of memory to ChatGPT is, in many aspects, a crucial and much-needed feature. However, it also comes with its own set of challenges. OpenAI's approach to implementing memory parallels the methods employed by other internet services in learning about users. Similar to how these services observe your interactions, track your searches, clicks, likes, and more, ChatGPT aims to build a profile of users over time through their engagements with the system.

However, this approach understandably raises concerns and discomfort for many people. Users are often cautious about their queries and messages being collected by OpenAI and utilized as training data to enhance the personalization of the bot. The concept of ChatGPT possessing knowledge about users is simultaneously intriguing and unsettling for many.

OpenAI emphasizes that users maintain control over ChatGPT's memory, assuring that the system is trained not to retain sensitive information, such as details about your health. You have the ability to inquire about what ChatGPT knows about you, instruct the bot to forget specific details, or manage memory in the new "Manage Memory" section of settings. OpenAI introduces Temporary Chat as a form of incognito mode, enabling quick conversations without affecting ChatGPT's memories of you. Users also have the option to disable memory entirely across their entire account.

By default, memory will be activated, and OpenAI plans to utilize memories to enhance its models in the future. Notably, companies using ChatGPT Enterprise and Teams won't have their data sent back to the models.

At present, the memory feature is in a testing phase, accessible to a "small portion" of users, as stated in the company's blog post introducing the feature. The potential integration of this feature into the standard interaction with ChatGPT raises considerations about how swiftly it could become a fundamental aspect of our interactions with the bot, both for better or for worse. As bots become more intelligent, their ability to quickly learn about us is a noteworthy development.