#Memory Leak - ChatGPT recalling conversations from other users

16 messages · Page 1 of 1 (latest)

shell bane
#

I was working fine with GPT4 all day and then out of no where, it started giving me answers to other people's conversations again. I thought this bug was fixed already. I am getting 100% unrelated answers now and they all seem to be random so I can only assume they belong to other users

#

I was working on code for a completely different application and it randomly started giving me code in an Python for an unrelated application involving Employees and Managers. When I tried regenerating the response, it was giving me code for a RandomForestClassifier for another unrelated application

#

Now its giving me a Fibonacci function.

#

I tried a fresh conversation and subsequent responses after the first one are all still giving me nonsense answers. GPT4 is unusable when it does not point to the right conversation. Please fix this data leak

shell bane
#

Same problem exists on default model. Seems to be system wide. None of this has anything to do with my original prompt

#

ChatGPT responding with answers to conversations from other users

#

Memory Leak - ChatGPT responding with answers to conversations from other users

#

Dev console does not appear to be a help

shell bane
#

Issue still persists. It is occurring whenever I am telling ChatGPT to continue a prompt

#

It's definitely looking at other users' conversations because I never gave it this prompt

#

None of these represent our conversation except the last 5 prompts. It keeps jumping around to other people's conversations

#

Memory Leak - ChatGPT recalling conversations from other users

shell bane
#

Okay so I am getting closer to figuring out what is going on. I tried using the same prompt on a new OpenAi account with no chat history whatsoever. And ChatGPT is functioning as expected on that account. On my primary account where I am a paying ChatGPT Plus subscriber, I am getting this unexpected behavior. Therefore it can only be one of two things or both:

  1. The bug is related to ChatGPT Plus subscribers
  2. The bug is related to reaching a hidden memory limit across all chat history

On my ChatGPT Plus account, I have a lot of chat history as I have been routinely using ChatGPT for productivity. New chats on that account experience the same bug. So if there is a maximum amount of memory between all chats and that limit is exceeded internally, it may be causing this issue. The issue is not relegated to one chat history. All new chat instances on this account now experience this issue as well which is why I believe it may be related to the total memory limit.

Steps to reproduce:

  1. Create a lot of chat history logs with dense chat histories on a ChatGPT Plus account (I have hundreds, where a lot of text input is approaching the maximum character limit)
  2. Keep testing long inputs until outputs match other users conversations

Additional information:

  1. Verify that last long input with inconsistent output does produces expected output on free account with no chat history
  2. Verify that last long input with inconsistent output does produces expected output on ChatGPT Plus account with no chat history
#

Memory Leak - ChatGPT Plus recalling conversations from other users

#

Welp that did not last long. As you can see, I was getting some code in my previous response and then after pressing it to continue with it's responses, It started pulling from a completely unrelated conversation. This is on the fresh account with no chat history outside of the current chat window. This seems to disprove my current thesis for how to reproduce the bug. For this account I gave it a large input and kept pressing it to continue annotating the output until I got this undefined behavior. It seems to occur across all models and all account types

#

Memory Leak - ChatGPT recalling conversations from other users