ChatGPT memory upgrade expands scope and can distort future answers

OpenAI has expanded ChatGPT's memory from a simple list of saved facts to a system that builds a profile from past chats, explicit instructions, personal constraints and implicit preferences. The upgrade, called Dreaming V3, synthesizes chat history in the background and can carry forward complex, multi-session context without an explicit prompt.

That synthesis can be misleading. The system has produced incorrect, specific summaries—claiming, for example, a user's Kasa plug was tied to Home Assistant when it was not—and old or irrelevant details from past conversations can persist unless users manually curate them.

Disabling memory stops background consolidation but does not remove stored memories or chat history, and deleting saved facts does not fully purge the model unless the underlying chat is deleted; safety-relevant context may also be retained in rare situations.

chatgpt, memory upgrade, dreaming v3, memory synthesis, chat history, saved facts, background consolidation, implicit preferences, home assistant, kasa plug