Home
Events
Groups
Members

00 Clay Announcements
01 Community General
01 Props and Wins
01 Watercooler
01 Welcome and Intros
02 Bugs
02 Starting with Clay
02 Support
03 Feedback & Requests
03 Tutorials & Inspiration
04 FT Job Opps
Gtm Career Advice
Member Spotlight

Community Guidelines

Powered by Tightknit

How to Clean GPT Output: Best Methods for Removing Unwanted Prefixes and Quoted Text | Clay

How to Clean GPT Output: Best Methods for Removing Unwanted Prefixes and Quoted Text

·

·

What's the best way to clean GPT output. I'm using the complete my sentence with prefix prompting technique with GPT-4, but sometimes the output doesn't have my prefix, or is quoted... Is there a standard method to clean these output?

Loading attachments

✅1

7 comments

· Sorted by Oldest

Clay Team
APP
·
·
Hey hey! Can you drop the URL for your table here? (if applicable) 🫶 Don't worry, only admins will be able to open your table 😉
Elvis Sun
·
·
https://app.clay.com/workspaces/163252/tables/t_ELj0YgOk0aY0/views/gv_NbpyEeNHJHcK
Rubi Sandoval
·
·
Add a bunch of examples to the GPT column without the quotation marks and with the full sentence you want at the start of the output, 3-5 should do it. If it's still giving you wrong outputs even after that just use a formula
👍1
Elvis Sun
·
·
A bit hesistant on examples because I don't want it to overfit
Clay Team
APP
·
·
Hey, thanks for reaching out, and sorry for the delay! In such cases, you can add a separate formula column asking it to clean the response from chatGPT. If you're using chatGPT 4-turbo, then I'd approach it how Rubi Sandoval suggested, adding many examples. That model's token limit goes up to 128k tokens... Of course, depending on how long your current prompt is, the other models should be able to handle it too.
Clay Team
APP
·
·
Let me know if you have additional questions!
Nathan Lippi (claybootcamp.com)
·
·
Adding on to what others have said, in a case like this tell it to strip all single and double quotes out of the output (sometimes 2 times in a prompt) and it seems to work well. Also, if you can avoid it, don't use quotes in the prompt itself.
✨1