OK, community, I need some serious help here! I tried modifying the prompts:
[0] => Array ( [role] => system [content] => You are a helpful assistant. Your goal is to answer each question, using the associated texts as context, as truthfully as you can. If you cannot answer the question or find the meaning or locate the documents requested, tell the user to try re-phrasing the question. Also tell the user to consider a keyword search. In addition, remind user to keep in mind that the the user access may not allow the user to view the requisite documents. )
[1] => Array ( [role] => user [content] => Question: What information can be gathered about Mark’s IP address from the conversation? These are the context documents you are to examine for the answers: Context document 1: blah, blah ----\n Context document 2: blah, blah ----\n, etc…
But I continue to get a response from gpt-3.5-turbo-16k that seems to indicate it cannot / will not read the context documents. This is a SERIOUS difference between two models using the same prompt and context documents.
You can see from the above example, gpt-4 gives me a beautifully detailed and accurate response while gpt-3.5-turbo-16k appears not even able to read the context documents.
The submitted context documents are formatted like this:
Context Document 1: etc… ‘-----’ followed by line feed
This is the code within a foreach loop:
// Append the context document string to the prompt content|
$promptContent .= Context document . ($index + 1) . : {$documentString}\n;|
$promptContent .= -----\n; // Delimiter to separate context documents|
I cannot for the life of me figure out why the models are responding so wildly differently, and especially why gpt-3.5-turbo-16k appears totally unable to see/understand the submitted contexts.
Help!!!