OpenAI has announced a major upgrade to ChatGPT memory, and while the company is presenting it as an improvement for personalization, the whole thing also feels a little unsettling. The new system is called “dreaming,” which sounds less like a software feature and more like the beginning of a sci-fi movie where the AI eventually locks the pod bay doors.
According to OpenAI, dreaming allows ChatGPT to automatically synthesize memories from conversations over time. Instead of only remembering details users explicitly save, the AI can now piece together habits, preferences, routines, projects, interests, and personal context from regular chats.
That means ChatGPT may remember your hobbies, travel history, dietary preferences, work projects, favorite products, sleeping habits, shopping tendencies, photography gear, or who knows what else.
OpenAI says this will make ChatGPT more useful because conversations become more personalized and contextual. To be fair, some of the examples shown by the company do look genuinely useful.
In one example, ChatGPT remembered a user’s exact underwater photography setup and recommended compatible camera accessories instead of generic advice. Another showed the AI planning a Singapore trip around a user’s preference for wildlife photography, strong hotel air conditioning, and quieter dining experiences.
That all sounds convenient, but it also sounds like OpenAI is slowly building a long-term behavioral profile of users.
The company says memories can update over time too. If you take a trip to Singapore, ChatGPT may eventually understand that the trip is over and stop recommending late-night takeout in another country weeks later. While that may solve the problem of stale memories, it also highlights how much ongoing context the AI is tracking in the background.
The bigger issue here is trust. OpenAI says users can review and manage what ChatGPT remembers, but most people probably do not realize how much personal context they casually reveal during everyday conversations with AI.
Folks increasingly treat ChatGPT like a therapist, doctor, financial advisor, relationship coach, coding assistant, travel planner, and search engine all at once. Over time, that creates a remarkably intimate portrait of someone’s life.
And unlike a notebook sitting in your kitchen drawer, this memory system exists inside a cloud-based AI platform operated by one of the most powerful technology companies on Earth. That reality alone should probably give some users pause, especially as AI companies continue racing toward deeper personalization and more emotionally aware interactions.
There are other risks too. AI memory systems can misunderstand context, infer things incorrectly, hold onto outdated assumptions, or over-personalize responses based on bad information.
Perhaps most importantly, these systems can make users trust the AI more simply because it feels familiar. Humans naturally trust things that remember them, and a chatbot that recalls your hobbies, favorite restaurants, work frustrations, health concerns, and family details starts feeling less like software and more like a relationship.
That emotional attachment could eventually become far more powerful than people realize.
OpenAI says the new memory system dramatically improves factual recall and preference tracking, and according to the company’s internal benchmarks, the latest version performs far better than previous memory systems from 2024 and 2025. Of course, these are OpenAI’s own numbers, so they should be viewed with at least some skepticism.
The company also says it reduced compute requirements enough to eventually bring these memory features to free users too, which means much more aggressive personalization could soon become standard across ChatGPT.
The uncomfortable reality is this: OpenAI no longer wants ChatGPT to simply answer questions. It wants the AI to know you over months and years.
Some people will love that. Others may eventually realize they handed an AI company a detailed running summary of their entire lives.
Support independent tech journalism
NERDS.xyz is independently owned and operated. If you enjoy my coverage of Linux, AI, hardware, cybersecurity, and tech culture, consider supporting the site on Ko-fi.
Support NERDS.xyz