The tool will show “updated memory” in the log, and a summary of the memory. But the actual “memory” doesn’t exist in the GUI, or in the printed set model context, and chatgpt can’t recall the information in a new thread (unless I enable chat history). My memory set context is also loaded with these duplicate weird “forget stuff” entries which makes me think there’s some kind of caching error, however others on reddit report the same memory saving issues without any odd set model context entries.
To be clear, some memories save, but it seems fairly random. I have personal history stuff I literally copied OUT of memory to test things, and they won’t go back in. So it doesn’t seem to be length or content related. But then other things seem to work fine. I’ve tried asking memories to be saved, using to=bio etc and even switching between all available models., and every time the GUI pops up with a clear memory save (suggesting that chatgpt does interact with the memory tool), but the memory cannot be accessed by the model, doesn’t appear in the Memories GUI, and doesn’t appear to be loaded in the set model context either.
I’ve tested this across various devices, browsers, and even on my old free account. Verified sporadic memories go through but probably 85% don’t. The same memory can fail then go through another time, so unlikely to be a filter issue (and as noted, many of the things I’m trying to add are memories chatgpt created for itself earlier).
