Question

Chat GPT dialogue trough Slack, how can it remember our previous talk?

  • 8 April 2024
  • 3 replies
  • 24 views

I am connecting Chat GPT to our Slack trough a Zap. I have set it up to reply in a Slack-thread for an on going dialogue. 

My problem is that it does not remember the first message, when continuing the dialogue. It understands to keep responding in the same thread, but i seems to have forgotten what we are talking about. 

Example:

Me “Is the earth flat?” 
GPT: “No its an oblate spheroid”
Me: “Is it a cube?”
GPT: “Please specify what is the object you are reffering to” 

I was thinking this is what the “Conversation ID” (Previously memory key?) is meant to be for, but i cant find anyone who actually managed to solve this. 

Do i need to find a way to feed it the previous conversation of that thread back in, everytime i want to continue the conversation? 

Providing an overview of my zap. Everything works here, it identifies what is a new message, and what is a message within a thread. It just needs to remember the previous messages within that same thread. 

Thank you!


 


3 replies

Userlevel 7
Badge +14

Hi @Kenny Berggren 

The Conversation ID is what will be used for ongoing memory.

Check the field description for guidance.

Make sure the original Slack Message is being sent to Assistant to have in the memory to reference.

 

Try removing the leading line return before the mapped variable.

 

Thank you for the reply! My problem is that the conversation ID does not seem to help in remembering the conversation. So what you are saying is that it needs to feed the original message back in to remember it? How do i do that? 

Lets say the dialogue is 5 messages back and forth in the same chat session, how do i set this up so that all 5 previous messages are being fed back as memory? 

Userlevel 7
Badge +11

Hi @Kenny Berggren 👋

Instead of the Ts field, you’d need to select the Thread Ts field, as it would have the timestamp of the original message not the timestamp that the thread was created from. The thing with that though is that I expect the Thread Ts field wouldn’t be there if the message that’s triggering the Zap hasn’t had a thread created for it yet. So in those cases you’d need to use the Ts field. 

To use the Ts field when no Thread Ts field isn’t available I’d suggest using a Formatter (Text > Default value) action to check whether a Thread Ts value exists and if not, use the Ts value instead. For example:
f63f5345827799f4c5e4e0245dd6993c.png

Then on the ChatGPT action, in the Conversation ID field you’d select the value that’s output by the Formatter action. For example:
1995da917ff29a05b33e2718d629f1bb.png


Can you give that a try and let us know if that then allows it to reference the previous messages in the conversation?

Reply