OpenAI is currently testing a new feature for its artificial intelligence-based chatbot called ChatGPT, which will save information after the end of the user’s interaction session with the digital assistant.

The relevant information is contained in a message that was published on the mentioned company’s blog last Tuesday, February 13. It is also known that the chatbot will be able to independently determine which fragments of a conversation with a user should be stored in digital memory.

The company plans to initially provide access to the new feature, currently undergoing corporate testing, to hundreds of thousands of those who interact with ChatGPT as part of a paid subscription, and those who communicate with the virtual assistant for free.

The new memory feature is another effort by OpenAI to make its globally popular chatbot more useful to people. Currently, the customer base of this digital product is 100 million people per week. The company is currently facing increasing competition. The number of firms offering users access to their chatbots based on artificial intelligence has grown. These digital products have a high level of cognitive development.

It is worth noting that saving the history of communication between a virtual assistant and a person can provoke a discussion about maintaining the confidentiality of information. The active development of advanced technologies in parallel with the advantages and impressive efficiency creates risks associated with cybersecurity. At the same time, developers are creating solutions to counteract the negative aspects of the functioning of AI systems. Currently, the problem of hacker activity also continues to be relevant. These criminals sometimes gain unauthorized access to their victims’ devices and blackmail them with threats of placing confidential data in the public access space. To counteract such attacks, it is enough for users to enter such a query in the Internet search system, such as, for example, how do I know if my phone camera is hacked? Information about the algorithms of cybercriminals’ actions will allow them to recognize unauthorized access.

Saving the history of the user’s interaction with the chatbot will help OpenAI improve the experience of each client and increase the level of engagement.

Joanne Jang, Head of the OpenAI Product Department, demonstrated the capabilities of the new feature by telling ChatGPT three different facts, one in each of the three different chats. She wrote that her daughter Lina, who will soon turn five, loves jellyfish and pink. In the fourth chat, Joanne Jang asked ChatGPT to create a greeting card for a child. The chatbot, using the DALL-E 3 image generator, generated a pink postcard with a picture of a jellyfish of the same color. The virtual assistant also wrote: Happy 5th birthday, Lina.

The company said that those users who will be granted access to the new feature will receive a notification. These consumers will be able to delete individual saved data, wipe all of it at once, or disable the option completely. Users can also ask the chatbot what it remembers to get a rundown of any information.

Joanne Jang says that ChatGPT already keeps its interlocutors from providing confidential information, including, for example, passwords and passport numbers. According to her, the new feature will not save such data.

The information stored in the ChatGPT user’s memory will be treated as ordinary conversational data. These information materials can be used to train artificial intelligence models if the user provides the appropriate permission.

Liam Fedus, a research scientist working on ChatGPT and GPT-4, one of the AI models that powers the chatbot, says that the memory function will be able to store a couple thousand tokens of written data. The OpenAI software breaks words down into segments consisting of several letters, known as tokens, for text processing.