OpenAI is launching a new memory architecture for ChatGPT called Dreaming. This system automatically synthesizes and updates conversation context without requiring explicit user instructions. The update aims to resolve previous issues regarding information staleness and accuracy.
A new memory summary page allows users to review, edit, and manage stored information. The Dreaming architecture is five times more compute-efficient than previous versions. This efficiency gain enables OpenAI to double memory capacity for paid subscribers.
The feature is currently rolling out to Plus and Pro users in the United States. OpenAI plans to expand the update to international markets and free users in the coming weeks.