OpenAI is equipping ChatGPT with a new feature named “Memory.” The tool allows users to train the language model on a more personal level, though users in Europe and Korea are currently left out.
The reason ChatGPT and other LLMs are so powerful right now is that they have been trained by millions of users, whether knowingly or unknowingly. The models will take what it has learned from various accounts and the broader internet to increase its knowledge base. Of course, individual users do not share the same persona as the entire internet.
ChatGPT is currently rolling out a feature named “Memory” that will keep a list of things to remember about the user. The feature is toggleable in the settings and will be available for the “Plus” tier, which excludes the free tier and those without an account.
In an example post, OpenAI demonstrated the tool. As the user tells the model something about their pets, ChatGPT notes it as a memory and displays a “memory updated” tag above the response. When that subject comes up again, the model recalls details from it’s previous memory and acts based on what it knows about the user.
Memories can be deleted.
OpenAI notes that the Memory feature will be rolling out to all ChatGPT Plus users, then goes on to specify that the rollout will not include Europe or Korea, likely due to regional regulations that may prohibit something like AI remembering specific aspects about existing entities – just a guess.
More on OpenAI:
FTC: We use income earning auto affiliate links. More.