OpenAI is upgrading its AI chatbot with a 'memory' feature, helping it store and recall
conversations
to deliver more personalised and
contextually relevant
responses.
Currently, in limited beta testing, the new
memory feature
allows
ChatGPT
to remember details users share about themselves, their preferences, and past conversations. It then applies this memory to answer questions, follow instructions, and make recommendations tailored to each user.
For example, if a user tells ChatGPT their daughter's name and upcoming birthday, the chatbot can store this information.
It can then retrieve it later to personalise a birthday card draft for their daughter automatically.
ChatGPT builds its memory by automatically flagging potentially useful details from conversations. Users can also explicitly tell the chatbot to remember or forget specific information. The goal is for ChatGPT to learn and evolve alongside each user over time.
Each custom GPT model will also have its memory. OpenAI provides an example of the Books GPT model, which, with memory turned on, can automatically remember which books you have already read and which genres you prefer. Other GPTs, including, Tutor Me could offer an improved course load once it knows what you already know, Kayak could go directly to your favourite airlines and hotels, and GymStreak could track your progress over time.
However, the feature raises concerns about ChatGPT gathering large amounts of personal data without proper transparency or control measures.
OpenAI says ChatGPT has been trained not to remember sensitive information proactively without explicit user direction. The company also states that users remain in control of ChatGPT's memory, with options to view, edit, or delete details the chatbot stores. OpenAI has also introduced a temporary chat mode that doesn't store any conversation information.
The ChatGPT-maker says that early feedback will guide it to expand the availability of the memory feature to more ChatGPT users in the future.
English (US) ·