OpenAI gives ChatGPT a memory: No more goldfish brain?

Artificial intelligence firm OpenAI is giving ChatGPT a “memory,” allowing the chatbot to remember past discussions and saving users from repeating themselves. 

In a Feb. 13 blog post, OpenAI announced that it would be rolling out the new memory feature to a “small portion” of ChatGPT premium and free users later this week, adding that plans for a broader rollout would be shared soon.

According to OpenAI, ChatGPT’s memory feature will improve over time as the chatbot interacts more with each user. The chatbot will be able to remember user preferences and previously shared details to provide more personalized and relevant responses as time goes on. 

The new memory function can also be adjusted in the settings, providing users with the ability to “explicitly tell it to remember something,” ask it what it remembers or even tell it to forget things completely.

Notably, the memory feature can also be switched off entirely. Temporary chats don’t use memory and won’t appear in history, and there are custom instructions that tell ChatGPT to remember things explicitly.

OpenAI stated that it “may use content” provided to ChatGPT, including memories, to train its large language models on an ongoing basis, but it added that data collection could be disabled.

The firm explained it was working to assess potential biases with the memory feature and would aim to have it avoid remembering sensitive information without explicit user direction.

We’re taking steps to assess and mitigate biases, and steer ChatGPT away from proactively remembering sensitive information, like your health details - unless you explicitly ask it to.

Additionally, OpenAI claimed that the memory feature could increase efficiency by learning work styles, formats and preferences for its premium users. Developers will also have the option to “have their own distinct memory” for their GPT builds.

In January, OpenAI rolled out an app store for subscribers of the pro version of ChatGPT, allowing users to discover and use customized AI chatbots known as GPTs for specific uses.

Meanwhile, one of OpenAI’s founding members, Andrej Karpathy, announced his departure from the Microsoft-backed firm.

In a Feb. 14 post on X, Kaparthy stated that he had left the firm the day before, adding that “nothing had happened” that influenced his decision to leave the company. He explained that he would be working on personal projects following his departure.