Instrumentation: User, System, Output

[harrisonchu]

In the current instrumentation approach its really hard to get the actual values from:

User Prompt
System Prompt
Output

It requires you to have system level knowledge of where these values reside within the object. In the case of OpenAI they are the first index of the array of the message field. This can also vary between providers.

[BEGIN DATA]
************
[Input Question, System message and Context to AI Assistant]:
{attributes.llm.input_messages.0.message.content}

{attributes.llm.input_messages.1.message.content}
************
[AI Assistant Answer]:
{attributes.llm.output_messages.0.message.content}
 ************
[END DATA]

I'd like to suggest we have an option in implementation to copy the parameters to well known fields. This makes the downstream accessibility much much easier, with the added burden of a slight bit of copying of data.

attributes.llm.user_prompt
attributes.llm.sys_prompt
attributes.llm.output

The other option is these are short cuts for the final data locations that will have to proliferate through our software everywhere.