
Tech • AI • Robotics
OpenAI is rolling out a new continuous memory feature called Dreaming, which could fundamentally transform how ChatGPT is used while raising concerns about the handling of personal data.
Until now, memory systems in conversational assistants were limited to a few static or poorly updated pieces of information. Dreaming introduces a dynamic memory capable of evolving over time, correcting inconsistencies and updating data based on context and past interactions.
The feature relies on ongoing information processing, even without direct prompts. Like a background task, the system analyzes daily exchanges to automatically enrich the user profile and refine future responses.
Thanks to this enhanced memory, responses become more relevant and tailored. Users may no longer need to repeat many instructions, as the tool can anticipate needs, preferences, and habits with greater accuracy.
Memory updates do not rely solely on conversations. The system can also use data from connected tools or contextual elements such as the date or planned activities, strengthening the consistency of responses.
A major concern involves the handling of sensitive information. Deleting a conversation no longer guarantees the removal of associated data, as it may already be integrated into the system’s internal memory.
The feature is automatically enabled for affected users. If disabled, a delay of up to 30 days may occur before stored personal data is actually deleted, raising questions about users’ real level of control.
Launched on June 4 for paid subscribers in the United States, the feature is expected to expand progressively to other regions and broader offerings, potentially including free users.
With Dreaming, OpenAI moves toward highly personalized assistants, while simultaneously prompting an important debate about the balance between technological performance and privacy.
Explain this