External Publication
Visit Post

Turn the fucking automatic memory creation/editing feature into an on/off setting

OpenAI Developer Community May 7, 2026
Source
I’m not saying remove memory entirely. Using existing saved memories to improve responses is useful. The problem is when ChatGPT creates new memories or modifies existing ones without the user explicitly saying “remember this.” That is not convenience. That is a user control problem. A user may want ChatGPT to reference existing saved memories. But that does not mean they want ChatGPT to automatically write new memories. These settings should be separate: Use saved memories: On Automatically create/edit memories: Off Why the hell is this not already an option? “Just turn memory off” is a bullshit answer. That also disables the existing memories the user actually wants to keep using. If chat history reference is already a separate feature, then automatic memory writing should absolutely be separately controllable too. A correction, complaint, temporary preference, or one-time instruction during a conversation should not be automatically turned into long-term memory. The fix is simple: Let ChatGPT read existing memories. Do not let it write new memories unless the user explicitly allows it. Do not let it edit existing memories unless the user explicitly requests it. Separate memory read and memory write permissions. Not having this basic on/off control while allowing automatic memory creation and modification is a fucking stupid design.

Discussion in the ATmosphere

Loading comments...