Skip to main content

Hello everyone!

As the title states, I’m making a FAQ chatbot with ChatGPT, but for some reason the request wont be trigger/filled sometimes.

The idea of this flow is to ask the prospect to make an appointment. A question is done (“Do you want to make an appointment”) then the response it’s send to chatGPT and it classifies the text as Positive/Negative/Null, then the response is used in a condition.

The flow:

 

The classification does work, it does give the correct context, but after trying out a few times (from 2 to 4 times) it gives a blank response with no apparent reason and the flow stops.

 

Working process:

(translate Q:want to make an appointment? / User:No / ChatGPT: Negative)

 

Fail process:

 (translate Q:want to make an appointment? / User:No / ChatGPT:  )

 

Things I’ve tried with no results:

  1. Validate the answer, if the answer is null, then come back to the chatGPT request to get a response (the flow stops instead of giving a blank answer, I think it gets in a endless loop and stops)
  2. Added a 2 seconds timer before and after the request
  3. Using different models (3.5 turbo and 3.5-turbo-16k-0613)

Note: The camp where the chatGPT response is stored, it’s deleted later on the same flow (I don’t think it affects it, but still)

 

 

Check Settings > Logs.

Most likely you’ll see an error there with indication of what’s wrong.

And I’ll bet that you’re exceeding your quota, because your OpenAI is on a low Tier.

If that’s the case, add credit to your account (I believe you have to add $50 nowadays to go up a Tier), and that should solve your problems.

We’ve done a bunch of these with @josemartinez.co , feel free to reach out if you need assistance.

@Gustavo Boregio man I love you! Just changed from free to tier 1 and its solved!

How do you choose which model version to use?

 

Hey @Dican happy to help!

I use the cheapest model that will get the job done!

I start playing in the playground to get my parameters right, then I copy the config over to Manychat.

I start with GPT3.5 Turbo, since it’s the cheapest. If I can get this working, then I use it! If I can’t get it to perform exactly the way I want, then I experiment with other models and GPT4 which is a lot more expensive.

We have a whole course about ChatGPT + Manychat in Spanish on cursos.superarmeonline.com

Reply


Community Terms & ConditionsCookie settings