• by Laurie Sullivan  @lauriesullivan, February 14, 2024

OpenAI is testing an option that lets ChatGPT retain information from one exchange to the next, the artificial intelligence startup said in a blog post Tuesday. 

The option is described similar to the way search engines and retail sites retain consented first-party information to use in future exchanges of information.

It is so similar to the way a search engine works that The Information, citing sources in the know, reports that OpenAI has been working to develop the next generation of an AI-powered internet search engine partially supported by technology from Microsoft Bing.

The user gives the chatbot the information and ChatGPT will retain it to become more helpful in the future. 

The company will initially make the feature available to hundreds of thousands of free and paid ChatGPT users, with plans to review feedback before rolling it out more widely.

Giving ChatGPT memory brings additional privacy and safety concerns and considerations, such as what type of information should be remembered and how it could be used.

The company said it will take steps to assess and mitigate bias, and will steer ChatGPT away from remembering sensitive information such as health and financial details, unless the users explicitly ask it to.

OpenAI provided several examples of how giving ChatGPT a memory can support future use.

For example, a user may mention that their toddler loves jellyfish, and when ChatGPT is asked to help create her birthday card, it suggests a jellyfish wearing a party hat.

Or if a kindergarten teacher with 25 students has a preference for creating 50-minute lessons with follow-up activities, ChatGPT will remember this when helping to create lesson plans.

Users have the option to turn off memory at any time, and while memory is off, the technology will not create, retain or use this memory option.

ChatGPT's memories evolve with each interaction and are not linked to specific conversations. Deleting a chat does not erase its memories. The entire memory must be deleted.

By using this option, the user gives permission to use the content provided to ChatGPT -- including memories -- to improve its models for everyone. But this option also can be switched off through the Data Controls in the platform.

OpenAI will not use the data to train content from ChatGPT Team and Enterprise customers -- for example, if someone asks about ad-campaign budgets or where they should invest money in a specific media. 

The company said it will continue to make available custom instructions that give users the ability to provide ChatGPT with direct guidance on what the user would like the technology to know about the person using it and how it should respond.