Experiencing Unexpected Model Responses from OpenAI API (GPT-4 Expected, GPT-3 Received)

My application uses data from web pages I upload, usually the same day. I created Mary Pason for this project and I have code to handle when the gpt fails to access websites.

It’s not possible for it make up the same information I am providing in web pages.

I don’t know what to tell you. Sorry.

from the FAQ:

https://help.openai.com/en/articles/6639781-do-the-openai-api-models-have-knowledge-of-current-events

another thread:

Hi @winninggrowthsystems
None of the models available on the OpenAI API have access to web browsing.

Hence the previous responses that you believe to be correct were just the model hallucinating and generating most likely tokens based on your prompt.

If you want to be able to browse web, you’ll have to write custom code to grab the contents of the url you enter.

This can be done either by prepopulating context within the prompt or using function calling.

I believe that version 0314 significantly outperforms the newer versions produced later, especially in few-shoot learning.But I can’t access 0314 anymore :joy:

2 Likes