Best answer

Text-davinci-003 Prompting Issue

  • 24 May 2023
  • 9 replies
  • 572 views

Userlevel 1

Hey guys!

 

I’m creating a chatbot through Zapier.

 

What I believe to be the final problem is prompting the OpenAI API to make it act as a certain persona.

 

I watched some videos on YouTube and I feel like I have a good understanding of what to tell the AI to make it more specific and to make it act as what I want. 

 

In fact the prompt I’ve created works perfectly in the OpenAI playground. The problem is the playground uses GPT 3.5, while OpenAI API still uses text-davinci-003. When I give davinci the prompt it doesn’t work as well, actually there’s a lot of issues. It looks like (and it makes sense to me) davinci is significantly less smart that GPT 3.5. I’ve been trying to tweak the prompt here and there to make it more clear to a “dumber” AI, but I’m struggling quite a bit.

 

Do you guys have any idea if GPT 3.5 will soon be integrated in the API so that I can use that? Or do you have some suggestions to improve the prompts for text-davinci-003?

 

Thanks everyone,

 

Alex

icon

Best answer by Danvers 25 May 2023, 12:57

View original

This post has been closed for comments. Please create a new post if you need help or have a question about this topic.

9 replies

Userlevel 7
Badge +12

Hi @AlexDLM!

You can use ChatGPT 3.5 turbo model if you use the ChatGPT integration and not the OpenAI integration. 

We can’t yet offer the more advanced models with the OpenAI integration because of the API that the two integrations use. The API is the code that allows apps to talk to other apps, in this case, what allows ChatGPT to talk to Zapier. We can access ChatGPT 3.5 on the ChatGPT integration but not on the OpenAI integration. 

Given that you’re building a chatbot, the conversation action should do what you need so switch to the ChatGPT app integration to access ChatGPT 3.5

 

​I hope that helps, please let us know if you have any questions!

 

Userlevel 1

Hi @AlexDLM!

You can use ChatGPT 3.5 turbo model if you use the ChatGPT integration and not the OpenAI integration. 

We can’t yet offer the more advanced models with the OpenAI integration because of the API that the two integrations use. The API is the code that allows apps to talk to other apps, in this case, what allows ChatGPT to talk to Zapier. We can access ChatGPT 3.5 on the ChatGPT integration but not on the OpenAI integration. 

Given that you’re building a chatbot, the conversation action should do what you need so switch to the ChatGPT app integration to access ChatGPT 3.5

 

​I hope that helps, please let us know if you have any questions!

 

Hi @Danvers ,

Thanks for the reply. I looked at the chatgpt integration and unfortunately it doesn’t let me update the prompt.

 

What I was looking at now was instead of using the “Send prompt” action in the “OpenAI” app, I could use the “Send API Request” action. This should allow me to use GPT 3.5 and let me give it a prompt at the same time. I’m looking now at the API Reference to try and make it work.

 

Let me know if you think this is a good idea!

Userlevel 7
Badge +12

Hey @AlexDLM

My understanding is that the OpenAI app uses an API called ‘Completions’ and that doesn’t support the gpt-3.5-turbo model. That model is only useable with the ChatGPT API. So, I’m not sure that you’ll be able to create a request that does what you need. I’m not an expert in APIs, and especially not the OpenAI one, so you can definitely take a look, but I don’t think it’s possible. 

The Make an API request action isn’t available for the ChatGPT integration, sorry about that. If you want to ask the team to add a feature request for that addition, could I ask you to please get in touch with the Support Team about that? (Using the Get Help form). 

I hope that’s all clear, and I’ll keep my fingers crossed that you can get things working the way you’d like with the OpenAI API. 

Userlevel 1

This is what I’m looking at: https://platform.openai.com/docs/api-reference/chat/create, and you can see the model compatibility here: https://platform.openai.com/docs/models/model-endpoint-compatibility. I’m no expert either so I’m not sure, but to me it looks like it should be doable.

Userlevel 1

This is what I’m looking at: https://platform.openai.com/docs/api-reference/chat/create, and you can see the model compatibility here: https://platform.openai.com/docs/models/model-endpoint-compatibility. I’m no expert either so I’m not sure, but to me it looks like it should be doable.

Actually you might be right, the chat API might not do the job for me. 

Userlevel 7
Badge +12

Hi @AlexDLM!

You beat me to it - I just found the endpoint compatibility doc. 

For clarity/in case someone else has a similar question:

ChatGPT is a product within the OpenAI suit of tools and it uses the /v1/chat/completions API endpoint, which is a subset (a part of) the entire API that OpenAI provides to access its tools. 

As you found, different API endpoints have access to different AI models (as shown on the Endpoint model Compatibility page)

 

The Zapier OpenAI integration uses a different endpoint to the Zapier ChatGPT integration because it does more than just chat (you can see the full list of actions available for the OpenAI integration here) which is why different models are available for them. 

 

The ChatGPT integration doesn’t currently have the API request action. You can use Webhooks by Zapier to send and receive information from the ChatGPT API, but it requires more work/understanding of thing like sending authorization in webhooks. If you’d like to use the Make an API request action with ChatGPT, please reach out to the Zapier support team using the Get Help form to submit a feature request for that. 

 

I hope that’s all clear and best of luck with your chatbot!

 

Userlevel 1

Do you think a new better model will be added to the OpenAI integration? Because the chatgpt integration doesn’t allow to give it a specific different prompt and I don’t think it ever will (it wouldn’t make sense).

Userlevel 7
Badge +12

From a Zapier point of view, we’re currently restricted by the models that are available at the different OpenAI endpoints, so I’m not sure if/when that will change. You could ask OpenAI if they are planning to add the different models to different endpoints? I’m sorry I can’t be of more help on that one. 

 

Userlevel 1

Thank you very much for your time!