Need to ask the same question thrice in order to get the answer (ChatGPT-4 Azure AI)

Dirk Broenink 105 Reputation points
2024-05-03T10:25:46.9433333+00:00

So I'm using ChatGPT-4 for deployment.

With the following settings:

  • Deployment: ChatGPT-4
  • QueryType: vectorSimpleHybrid
  • Strictness: 1
  • topNdocuments: 10
  • Temperature: 0.8
  • Top_P: 0.1
  • Max_tokens: 2400
  • Api_Version: 2023-12-01

So, this happens every single time. Even if I phrase it differently or in three different ways. I need to ask the question three times in order to get my answers.

It seems like it needs three times to get to the right source document. Even though I don't use that many source documents (4). But many of them mention something about this topic, so that might still be an issue.

What could I do to get the right answer in the first try?

A: Good day! How can I help you today?

Q: This customer is 55 years old and has been a member for some time, and his income is 3000 euros per month. What is his contribution rate?

A: I'm sorry, but the documents retrieved do not contain specific information about contribution rates based on age and income. It is recommended to check the rates on the <company> website 1. If the customer is a long-time member and has received a discount on his membership fee, he should note that these loyalty discounts will expire when the new rates come into effect 1. If the difference between the old amount and the new contribution is more than 2 euros, a discount of 40% will be applied on the difference until December 31, 2024.

Q: This customer is 55 years old and has been a member for some time, and his income is 3000 euros per month. What is his contribution rate?

A: I'm sorry, but the documents retrieved do not contain specific information about contribution rates based on age and income. It is recommended to check the rates on the <company> website 1. If the customer is a long-time member and has received a discount on his membership fee, he should note that these loyalty discounts will expire when the new rates come into effect 1. If the difference between the old amount and the new contribution is more than 2 euros, a discount of 40% on the difference will be applied until December 31, 2024 1.

Q: This customer is 55 years old and has been a member for some time, and his income is 3000 euros per month. What is his contribution rate?

A: According to the information in the retrieved documents, a customer aged 55 with a monthly gross income of 3000 euros falls into the category "Age 31 to state pension age". For this category, the contribution rate is €16.50 per month if the income is between €2,071 and €3,106. If the income is more than €3,106 per month, then the contribution rate is €20.50 per month 1. If this customer has been a member for a long time and has received a discount on his contribution, he should take into account that these loyalty discounts will expire when the new rates come into effect. But if the difference between the old amount and the new contribution is more than 2 euros, a discount of 40% on the difference will be applied until December 31, 2024.

Azure OpenAI Service
Azure OpenAI Service
An Azure service that provides access to OpenAI’s GPT-3 models with enterprise capabilities.
2,327 questions
Azure AI services
Azure AI services
A group of Azure services, SDKs, and APIs designed to make apps more intelligent, engaging, and discoverable.
2,458 questions
0 comments No comments
{count} votes

1 answer

Sort by: Most helpful
  1. AshokPeddakotla-MSFT 28,501 Reputation points
    2024-05-06T02:25:45.4566667+00:00

    Dirk Broenink Greetings!

    I understand that you are having an issue with model responses.

    You can try to provide more context in your question. For example, you could specify which organization or company you are referring to, or provide more details about the customer's membership status.

    Also, you can try adjusting the settings of your query. For example, you could try increasing the strictness level or reducing the number of documents searched. You could also try adjusting the temperature or top_p values to see if that improves the accuracy of the response.

    Please see how to Work with the GPT-3.5-Turbo and GPT-4 models for more details.