The model is mixing up the input data

API
, ,
1

Hello,

I’m passing some data in the form of key-value pairs to GPT-4o along with the prompt. The problem is that sometimes if I’m passing a pair like “Food Addiction: High”, I see in the response that it considers it as “Low”. I cannot say it’s a problem in my prompt because this happens for a pair or 2 in 30 key-value pairs. However, it’s important to me to have correct predictions.

How can I reduce such hallucinations?

Thanks in advance!

1 Like
2

Make sure you describe each parameter for him to understand how it works and how to use it. Make descriptions and steps how to use them.

Treat GPT is like 5 years old when you give him instructions and will make amazing things if you reinforce him with detailed, specific information to get the desired output.