OpenAI gives ChatGPT a better 'memory'
OpenAI gives ChatGPT better memory.
The San Francisco artificial intelligence startup said Tuesday it is releasing a new version of its chatbot that remembers what users say so it can use that information in future chats.
If a user mentions a daughter, Lina, who is about to turn 5, likes the color pink, and likes jellyfish, for example, ChatGPT can store this information and retrieve it as needed. When the same user asks the bot to “make a birthday card for my daughter,” it can generate a card with pink jellyfish that says “Happy 5th birthday, Lina!”
With this new technology, OpenAI continues to transform ChatGPT into an automated digital assistant that can compete with existing services such as Apple's Siri or Amazon's Alexa. Last year, the company allowed users add instructions and personal preferences, such as details about their job or the size of their family, which the chatbot must take into account during each conversation. Now ChatGPT can draw on a much broader and more detailed set of information.
“We think the most useful assistants are the ones that evolve with you – and keep up with you,” says Joanne Jang, an OpenAI product lead who oversees the memory project.
Although ChatGPT can now remember past conversations, it can still make mistakes – just like humans can. When a user asks ChatGPT to make Lina a birthday card, the chatbot may create one with a subtle typo such as “Haippy 5th birthday! Lina!”
The company will first deliver the new technology to a limited number of users. It will be available to people who use the free version of ChatGPT, as well as to those who subscribe to ChatGPT Plus, a more advanced service that costs $20 per month.
OpenAI also introduces so-called temporary chats on Tuesday, where conversations and memories are not saved.
ChatGPT has been offering a limited form of memory for some time. When users chatted with the bot, responses were based on what they said earlier in the same conversion. Now the bot can draw on information from previous conversations.
(The New York Times sued OpenAI and its partner Microsoft in December for copyright infringement of news content related to AI systems.)
The bot builds this memory by automatically identifying and storing information that could be useful in the future. “We rely on the model to decide what is or isn't relevant,” says OpenAI researcher Liam Fedus, referring to the AI technology underlying ChatGPT.
Users can tell the bot to remember something specific from their conversation, ask what is already stored in its memory, tell the chatbot to forget certain information, or disable the memory completely.
By default, OpenAI has recorded full ChatGPT conversations and used them to train future versions of the chatbot. OpenAI said it removed personally identifiable information from conversations used to train its technology. And users can choose to remove their conversations from OpenAI's training data entirely.
But creating and storing a separate list of personal memories that can be brought up by the chatbot in conversations can raise privacy concerns. The company argued that what it was doing was not that different from the way search engines and browsers kept their users' Internet history.