ChatGPT Memory Compression Bug: Reducing Memory Actually Increases Usage

問題描述(Issue Description)
我嘗試讓 ChatGPT 壓縮記憶,以減少長期記憶的容量消耗。然而,每當 AI 簡化內容、刪除不必要的細節後,記憶容量的顯示數值 反而上升,而不是減少。例如:

  • 初始記憶顯示 97%
  • 第一次壓縮後變成 98%
  • 進一步壓縮後變成 100%

這與預期行為不符。按理來說,刪減記憶應該讓佔用的容量降低,而不是增加。這是否是一個 Bug,還是記憶系統有不同的計算方式?

希望改進的地方(Suggested Improvements):

  1. 修正記憶容量計算方式,確保刪減內容後能正確反映變化。
  2. 增加真正的「記憶壓縮」功能,讓 AI 內部使用更高效的記錄方式,以減少容量佔用。
  3. 允許使用者手動刪除特定記憶,以便更精確管理長期記憶。

請問有其他人遇到相同的問題嗎?這是預期行為還是系統異常?希望能得到官方團隊的解釋或修正建議。

(English Translation for Devs):
I attempted to reduce ChatGPT’s long-term memory usage by simplifying stored content. However, after each optimization attempt, the displayed memory percentage increased instead of decreasing. For example:

  • Initial memory usage: 97%
  • After the first compression: 98%
  • Further compression: 100%

This behavior seems counterintuitive. Normally, removing stored content should decrease memory usage, not increase it. Is this a bug or an intended design of the memory system?

Suggestions for Improvement:

  1. Fix memory usage calculation to correctly reflect reductions.
  2. Implement true “memory compression” to allow more efficient internal storage.
  3. Allow users to manually delete specific memories for better control over long-term storage.

Has anyone else encountered this issue? Is there any official clarification or fix available?