Personalization and model improvement are not the same decision: Memory concerns what the system can remember, while the cited statements do not establish that it changes or improves the model. When Memory is enabled, the guidance says the system can remember relevant preferences and details from a user's chats and other available sources. The guidance also says that app permissions are distinct from settings that apply generally to all conversations.
Data controls: why the distinction matters
Memory is described in terms of remembering relevant preferences and details, which is the personalization side of the question. App permissions have a separate boundary: the guidance keeps them apart from settings that apply generally to all conversations. That separation means an app-level permission should not be treated as evidence of a general conversation setting, and Memory should not be treated as evidence of model training or improvement.
How to check the settings
A careful check should keep the questions separate:
- Memory: Whether it is enabled should be confirmed, along with the current explanation of the sources it can use. The cited wording includes chats and “other available sources,” but it does not enumerate every possible source.
- App permissions: Each app permission's current status should be reviewed, then compared with controls described as applying generally to all conversations. A change in an app permission should not be assumed to change a generally applicable conversation setting.
- Model improvement: This should be treated as a separate question. Neither cited statement describes a model-improvement setting or its effect, so a current, explicit explanation is needed before making that claim.
What readers must still confirm
The two cited statements leave several practical questions open:
- whether Memory is enabled for the relevant user;
- what “other available sources” covers in the current context;
- what a particular app permission allows or changes; and
- whether any setting affects model training or improvement.
The current controls and documentation should answer these questions directly. Until they do, remembered personalization should not be described as model improvement, and an app permission should not be treated as proof of a general conversation setting.