ChatGPT responses are getting trimmed more and more as I test it out with the same prompt.
Is there a way to fix this?
This is the message it shows up at the end of the response:
history_context:
was_trimmed: true
Answer
ChatGPT answer getting trimmed more every test action
- Beginner
Best answer by Danvers
Hi
I did some digging with our developers and learned more about the history_context.was_trimmed field.
When you use ChatGPT, Zapier stores the responses so they can be used in future prompts/zaps. Think of the whole conversation as a google doc, and there is a fixed sized window that captures what we can send to ChatGPT, like a limited field of view. That window has to include the most recent message, so it's pinned to the "bottom" of the doc. If the top of the window excludes some of the messages at the "top" of the doc (the earliest messages), then the was_trimmed is true.
history_context.was_trimmed is true when Memory Key is used but we needed to pop some of the oldest memorized messages to stay under the token limit.
I hope that's clear, please let us know if you have any questions!
This post has been closed for comments. Please create a new post if you need help or have a question about this topic.
+12- Zapier Staff
Hi
Can I check something? Do you mean that the messages are getting trimmed more frequently, or that even if the messages are shorter, they are still getting trimmed (ie they are getting shorter and shorter)?
This may be to do with the number of tokens used in the test and monthly billing limits from ChatGPT, but I’m not 100% sure about that. I’m going to check in with some of the ChatGPT experts we have at Zapier and will get back to you if I have some more info!
Thanks!
+12- Zapier Staff
- Answer
Hi
I did some digging with our developers and learned more about the history_context.was_trimmed field.
When you use ChatGPT, Zapier stores the responses so they can be used in future prompts/zaps. Think of the whole conversation as a google doc, and there is a fixed sized window that captures what we can send to ChatGPT, like a limited field of view. That window has to include the most recent message, so it's pinned to the "bottom" of the doc. If the top of the window excludes some of the messages at the "top" of the doc (the earliest messages), then the was_trimmed is true.
history_context.was_trimmed is true when Memory Key is used but we needed to pop some of the oldest memorized messages to stay under the token limit.
I hope that's clear, please let us know if you have any questions!
- Beginner
Hey
I am testing out a prompt with some questions that I am answering. So the prompt is the same every time, I only adjust a few things. And every time I try it, it comes out shorter and shorter.
I think this memory problem is also influencing my tests since I get the same exact response every time, and my goal is to get a different response since the response I am getting in not accurate.
How do I remove this Memory Key? I want this process to be new every single time with nothing in its memory to affect it.
Thank you so much for your time!
+6- Zapier Staff
Hi
I’m jumping in to see if I can help!
Unfortunately, there isn’t a way to remove the Memory Key. You mentioned:
I want this process to be new every single time with nothing in its memory to affect it.
How about we try sending a unique memory key for each time the Zap runs? To do this, you can add Text in Formatter by Zapier. Just select "Spreadsheet Style Formula" and enter RAND() into the formula field. This will ensure that each time the Zap runs, a different memory key is used. We've discovered that rotating the memory key resolves the issue, as it's a known limitation with our ChatGPT integration.
Please keep us posted! 😊
Enter your E-mail address. We'll send you an e-mail with instructions to reset your password.