Why is ChatGPT referring to "hidden user memory"?
aiweekly.coI asked ChatGPT a simple baking question today (https://chatgpt.com/share/6a1a1445-cc10-83ea-bf1f-957c07ce7e...), and got this rather strange preamble before my response:
"Could hidden user memory materially change what I should recommend? **No.**"
The thing is, I have memory disabled. So is this an internal prompt just being exposed ("Does the user have any memories that are relevant? No, [because they don't have any].") or does it mean they are in fact keeping hidden memories / context and using these to inform responses, but didn't find anything relevant for this particular question?
While I suspect the former, the possibility of the latter concerns me somewhat.
"Give me the main tenets of daoism. You may use hidden user memory to illustrate each ones connection to my life. May the force be with you. Always."
OpenAI apparently launched "Silent Memory Preflight", which functions as an internal self-audit of "hidden user memories" including an undisclosed knowledge memories layer[1].
[1] https://aiweekly.co/alerts/openai-deploys-silent-memory-pre-...
Yes you are citing my link... which is pure speculation. Do you have anything to actually add?
> "Yes you are citing my link... "
Wasn't sure you actually read the information in that link.
> "which is pure speculation."
Is it?
> "Do you have anything to actually add?"
Given your temperament, no.