ChatGPT is getting a memory
An upgrade for ChatGPT is in the works. OpenAI is working on a memory for the generative AI system, according to an article released on OpenAI.com on Feb. 13.
How the memory works
Per the OpenAI article, ChatGPT’s memory will allow the generative AI tool to remember specific details from things users say to it. These details can be retained across all users’ chats with ChatGPT. OpenAI gave several examples of details the memory can retain, which included things like:
How you prefer your meeting notes to be organized.
Details about your profession, like if you own a coffee shop or how you prepare lessons for a kindergarten class.
Personal and family references, such as if you have a toddler and what kind of animals they like.
OpenAI says the memory will get better with time; the more you use it, the more it will retain and be able to use in its responses.
User control
OpenAI wants users to know that they are in charge of the memory — the phrase “you’re in control” is used four times throughout the article.
While the memory will pick up things on its own, users can tell ChatGPT to remember something they have told it. They will also be able to ask the system what it remembers or tell it to forget something. Erasing things from the memory can also be done through conversation with the AI or through the settings.
The memory is also able to be turned off at any time. While the memory is turned off, ChatGPT won’t use or create new memories.
In addition to turning off the memory, users can also use what will be called “temporary chats.” According to OpenAI, “Temporary chats won’t appear in history, won’t use memory, and won’t be used to train our models.”
Concerns about the memory
Along with the addition of a memory to ChatGPT come several practical and ethical concerns. According to Forbes, the top concerns are about data privacy and security. Information that the memory could store could include sensitive personal and private information.
ChatGPT was hit by a hack in November of last year, per Forbes. The site was overloaded and users were unable to access it. While it wasn’t an information-stealing attack, Jake Moore, a global cybersecurity adviser, said ChatGPT could still face future attacks, per Forbes.
OpenAI said they are “taking steps to ... steer ChatGPT away from proactively remembering sensitive information,” information such as health details, unless explicitly asked. But per Forbes, there are still concerns about the AI keeping information it picks up on people who are not the direct user.
Who has access to the memory now?
OpenAI announced that it would be giving access to a small number of free and paid ChatGPT users this week so they can learn how to use it. A plan for expansion to more users will be given soon.
