Introduction
Repeating preferences and having to explain projects every time you start a chat is a common annoyance. OpenAI has introduced an update that promises ChatGPT will remember more and forget less, to avoid those repetitions.
What happened
On June 4, 2026, OpenAI began deploying a new memory architecture based on its system called "dreaming". The company says the update improves the freshness, continuity, and relevance of what the assistant remembers. It is now available for Plus and Pro users in the US and will gradually roll out to other countries and Free and Go users in the coming weeks. Dreaming is not a command users activate in the conversation, but a background process that reviews chats and synthesizes the essential into a visible memory state for the user. Generated memories can be viewed and corrected from the memory summary page.
Three practical changes for the user
Fewer repetitions when starting a conversation
ChatGPT will be able to use previous information about your projects or team to avoid having to re-explain details. This speeds up tasks like asking for technical recommendations or picking up a plan started earlier.
Preferences respected more consistently
Preferences and restrictions — for example, diet or response style — will be automatically applied in new conversations. The system aims to follow these preferences without you having to remind it each time.
More current and useful recommendations
Memories are updated over time: completed plans or finished trips move to a historical state, so suggestions (places, times, availability) adjust to the most recent reality.
How it works in one image
Imagine an assistant that takes notes of your conversations, organizes them, and creates a summary card you can review. Dreaming acts like that archivist: it reviews multiple chats, extracts relevant information, and presents it for the assistant to use or for you to modify.
What is not resolved
OpenAI acknowledges that previous versions could become outdated or incomplete, and that dreaming improved that, but it does not claim the memory is infallible. The feature is being rolled out in phases and will continue to be improved; manual review and control of memories remain necessary.
Closing
The update reinforces the trend toward more proactive and personalized assistants. At the same time, it underscores the need for clear tools to review and correct what the AI remembers.
Why it matters
- Business: reduces operational costs and makes it easier to offer memory to free users.
- Users: less repetition, more relevant and personalized responses.
- Industry: marks progress in scalable memory for conversational assistants.
🎧 Listen to the AI news summary here
Transparency Note
This article was written with support from AI tools. The source was approved by Diario Occidente and the final content was reviewed by a member of the editorial team.
