ChatGPT Now Remembers Your Preferences: How Its New Memory Works

What happened?

OpenAI has launched a new memory feature for ChatGPT, internally called 'Dreaming', that allows the model to remember explicit information provided by the user across different conversations. According to OpenAI's official blog, this capability aims to 'keep context fresh and relevant' to deliver more useful and personalized responses. Users can instruct ChatGPT to remember details such as their name, formatting preferences, interests, or recurring information, and the model will store them for future interactions. The feature is rolling out gradually, starting with a small group of ChatGPT Plus users, and is expected to reach all paid subscribers in the coming weeks, followed by free users. This launch represents a significant shift in ChatGPT's architecture, which previously operated without persistent memory between sessions.

Why is it important?

Until now, each conversation with ChatGPT started from scratch, forcing users to repeat information and limiting long-term coherence. With this improvement, the assistant can build a dynamic user profile, enhancing efficiency and experience. For businesses, this means more adaptive chatbots that can remember customer preferences, purchase history, or project settings. For individual users, interaction becomes more natural, similar to talking to a human assistant who remembers personal details. OpenAI has implemented privacy controls: users can view, edit, or delete what ChatGPT remembers, and can disable the feature entirely. This responds to growing demands for transparency in AI, especially after previous controversies over data usage. The company states that memories are not used to train models unless the user explicitly opts to share training data, a measure aimed at balancing personalization and privacy.

Historically, the lack of memory has been a major criticism of AI chatbots. From early rule-based systems to current language models, the inability to remember information between sessions has limited their practical utility. Competitors like Google with Bard (now Gemini) and Anthropic with Claude have also explored persistent memories, but none have implemented a solution as integrated as OpenAI's. For example, Claude allows 'projects' with persistent context but does not automatically remember preferences across projects. This update marks a milestone in the personalization race, positioning ChatGPT as a leader in creating truly contextual assistants.

Consequences and challenges

Persistent memory opens the door to applications like personal assistants that remember appointments, dietary preferences, or writing style. For instance, a user could ask ChatGPT to remember their birthday or that they always prefer responses in list format. In a business setting, a sales team could configure ChatGPT to remember client names and interests, streamlining communication. However, it also introduces risks: if a user accidentally shares sensitive information, it could be stored. OpenAI claims data is used only to improve the experience and not to train models (unless the user opts to share training data). The feature is rolling out gradually on ChatGPT Plus and soon on free versions. Competitors like Google (with Bard) and Anthropic (Claude) are also exploring persistent memories, so this update marks a milestone in the personalization race.

Security challenges are notable. Although OpenAI has implemented controls, the possibility of sensitive information being stored without the user's knowledge is a risk. Compared to past events, such as the Samsung data leak incident in 2023 where employees shared proprietary code with ChatGPT, this feature could exacerbate risks if not managed properly. OpenAI recommends not sharing sensitive information through chat, but responsibility lies with the user. Additionally, the feature could be exploited by malicious actors to extract stored information through social engineering. Transparency in data management will be key to maintaining user trust.

What readers should know

• Memory is optional and can be managed from settings.
• ChatGPT will not remember anything unless the user explicitly asks it to.
• Stored data can be deleted at any time.
• This feature is not immediately available in all regions.
• OpenAI recommends not sharing sensitive information through chat.
• Memories are stored securely and not shared with third parties without consent.
• Users can view a summary of what ChatGPT remembers in account settings.

'ChatGPT's memory is a step toward assistants that truly understand our long-term needs, but also a reminder that privacy must be a priority.' — Analyst at TheVortiq

In summary, ChatGPT's memory represents a significant advance in conversational AI personalization, but raises important questions about privacy and security. As the feature expands, it will be crucial to monitor how users interact with it and what additional measures OpenAI takes to protect data. In a market where differentiation increasingly depends on user experience, persistent memory could be the decisive factor that solidifies ChatGPT as the preferred assistant, provided it is handled responsibly.