I would like to suggest a feature that allows ChatGPT to access message timestamps, or at least relative time information, within a conversation.
Currently, the model appears to have strong sequence memory, but limited time memory.
In long-term conversations, users often refer to events using expressions such as:
-
“today”
-
“yesterday”
-
“a few days ago”
-
“last week”
-
“just now”
The model can usually understand the order of events, but it cannot accurately determine when those events actually happened.
As a result, events that occurred several days ago may sometimes be treated as if they happened today, simply because they are still present in the conversation history.
Example:
A user shares photos of clouds, birds, and cats several days ago.
A few days later, the user says:
“I got drunk today.”
Because the model only sees conversation order and not actual timestamps, it may incorrectly associate those older events with the current day.
In other words:
The model has Order Memory, but lacks Time Memory.
Suggestions:
-
Provide message timestamps to the model.
-
Alternatively, provide relative time information (e.g., “3 days ago”, “1 week ago”).
-
Offer an optional “Time-Aware Mode” for users who maintain long-term conversations, journals, research logs, project tracking, or personal life records.
This would significantly improve:
-
Long-term companionship
-
Personal journaling
-
Research and experiment logs
-
Project management
-
Health and habit tracking
Understanding what happened is important.
Understanding when it happened is equally important.