I removed “PDF_HTML” and input the function inputs instead > updated the prompt to direct the chatbot to generate a PDF after it finishes generating the output for the user.
@ned_rvth yes had some issues as well. As I want more flexibility on how the pdf’s are displayed and being able to quickly change the format, I have used Make in this case too (sorry I know I repeat myself)!
This seems to be an LLM issue. Depending on how you’ve prompted them, they can be quite adamant.
There is indeed differences between LLMs. Claude and GPT can behave differently. For example, I’ve noticed that when I connect a DALL-E action, Claude can be quite resistant to the idea that it can use DALL-E. Fortunately, you can always fix this issues with prompting.
I would revisit the prompt and explain very clearly that it can generate PDFs and how it should be done.
I managed to resolve the issues. There was no need for a Make integration. I switched from going back and forth with Claud 3.5 (using a dedicated project) to using OpenAI playground with the latest version of GPT 4o. Rewrote and organized both the code and the prompt, even added a user input to personalize the PDF output.
Notes and Tips:
*Claude seems to be stubborn or even clueless on how to work with wrapped GPTs. Use OpenAI’s playground to diagnose issues a third of the time.
Click “Format as JSON’ to view the code and make edits”