Crafting the most optimal Prompt to OpenAI from Slack

Hi there, I’ve been working on a workflow that triggers from Slack and then to the OpenAI API. The replies are coming back, however, they are not very relevant to the original trigger message.

I have the Playground within OpenAI set up and working perfectly. The responses are completely relevant and the context is maintained. I have tried just copying and pasting the entire thing into the Prompt section of Pipedream, but the replies then are sort of relevant, but they seem to be getting cut off.

I have adjusted the Bot model, Max Tokens, Temperature, Presence Penalty, etc exactly to what is set within the OpenAI Playground, so I’m wondering if anyone else might have any idea what the issue can be?

Thank you so much.

Hi @creatiq

Is it possible that your prompts aren’t being passed to your model correctly?

Can you share your step configuration or code that’s sending this data to OpenAI?

Hi @pierce, yes that is absolutely possible.

I recorded a Loom video which might give you better insight. Here it is here: Loom | Free Screen & Video Recording Software | Loom

Please let me know if I can answer any questions.

Thank you very much!

A little more progress since posting the video…

I realised that I had pasted the raw text from the Playground, as opposed to the Prompt from the Code view lol. So now, the return replies are more relevant and complete.

However, the reply that comes through is only what I specify in the prompt. There’s no dialogue.

For example, within the Playground, I had developed an entire persona called Athena with a series of information about our company. And then when I copy the code view of the prompt from OpenAI Playground and paste into the prompt area within Pipedream, the reply message is only related to the last part of the conversation.

Should I be studying more of OpenAI’s completion docs, or more of something on Pipedream? Or would you recommend that I find a Pipedream Partner?


Thank you again.

Hi @creatiq

Glad you were able to sort out the prompt issue.

Our support is limited to Pipedream specific questions, like setting up workflows, wiring up data, integrating accounts, etc.

You may find better support contacting a Partner who can consult on this API specifically.

Or you can read deeper into OpenAI’s documentation to learn how to keep a “session” in place.

Ok got it. Thank you very much for your assistance, @pierce!

1 Like