OpenAI has updated its ChatGPT memory system to create coherent, narrative-based profiles of users, organized into categories like work, hobbies, and travel. The new system automatically builds and updates these profiles in the background based on conversations, eliminating the need for explicit prompts. This change aims to address the issue of stale memories by ensuring the AI adapts to users' evolving preferences and contexts.
The improvements have led to significant gains in performance metrics. Fact retrieval success rates have increased from 41.5 percent in 2024 to 82.8 percent currently. The system's ability to factor in personal preferences has also improved, rising from 31.4 to 71.3 percent. Freshness of information has seen the largest jump, climbing from 52.2 percent last year to 75.1 percent now. OpenAI highlights these enhancements through examples, such as tailoring travel recommendations based on a user's interests in wildlife photography and quiet dining.
The update is initially available to Plus and Pro users in the US, with free users receiving it in the coming weeks. OpenAI attributes this progress to reduced computational requirements, which have been cut by a factor of five. The original memory feature, launched in April 2024, only saved information triggered by user commands, while the new system processes chat history automatically.
Source: thedecoder
