Dreaming: Better memory for a more helpful ChatGPT
New memory synthesis capabilities are being rolled out for ChatGPT, enhancing its ability to retain user preferences and context across conversations. This system, built on "dreaming" technology, aims to provide fresher, more relevant interactions by learning from chat history. It allows for a more personalized and efficient user experience, available for Plus and Pro users initially, with wider release planned soon.
ChatGPT is rolling out an advanced memory synthesis system to improve user experience. This system addresses previous limitations like staleness and scalability in handling memory across millions of users and extended timeframes. The core idea is to allow ChatGPT to learn user preferences, projects, and constraints, ensuring future conversations build upon shared context rather than starting anew. This significantly enhances ChatGPT's usefulness by fostering deeper understanding and enabling it to assist users in achieving long-term goals.
This new system, built upon a concept called "dreaming," represents a substantial upgrade from earlier "saved memories" functionality. While saved memories relied on explicit user instructions and often became outdated, "dreaming" operates in the background, automatically curating memories from chat history. This allows ChatGPT to continuously synthesize and provide the freshest, most relevant context without requiring direct user input. It also facilitates the inclusion of naturally occurring conversational context, leading to more fluid and intuitive interactions.
Users can review the memories synthesized by "dreaming" through a memory summary page. This feature enables quick access to highlights of what ChatGPT knows, allows users to update their information, and provides control over topics ChatGPT should bring up. This transparent management empowers users to refine their personalized ChatGPT experience.
The update is currently available to Plus and Pro users in the US, with plans for a broader rollout to additional countries and Free and Go users in the coming weeks. This phased release ensures a smooth transition and continuous improvement of this critical feature.
Related articles
Build real agentic apps using CUGA: two dozen working examples on a lightweight harness
CUGA, IBM's open-source Agent Harness, simplifies building agentic applications by handling infrastructure, allowing developers to focus on tools and prompts. It offers pre-assembled components for planning, execution, and state management, significantly reducing development time. CUGA has topped agent benchmarks like AppWorld and WebArena.
OpenAI launches new initiative to help find and patch open source bugs
OpenAI has launched "Patch the Planet," a new initiative in partnership with cybersecurity firm Trail of Bits, to enhance the security of open-source projects. This program aims to assist maintainers in identifying and patching bugs, utilizing OpenAI's AI-powered security tools while reducing the burden on project teams.
PP-OCRv6 on Hugging Face: 50-Language OCR from 1.5M to 34.5M Parameters
Baidu has released PP-OCRv6, an advanced optical character recognition (OCR) model supporting 50 languages. Available on Hugging Face, this version significantly improves accuracy and efficiency across various parameter sizes, from 1.5 million to 34.5 million, marking a substantial leap in multilingual OCR technology.
