RikY2 Prompting

RikY2 is unique in its ability to take in a core prompt and two sub-prompts, or follow-up prompts, when generating a response. This allows you to break down a complex or multi-faceted prompt into smaller, separate prompts. Because of the model’s architecture, this prompting style yields better, more accurate results.

Follow-up prompting can be accomplished by submitting a nested array (an array of strings within an array):

{
  "question": [
      "What is the name of the patient?", // This is a standalone question 
    [ 
      "What is the first medication listed under Section C - Medical Information?", // This is a nested array 
      "What is its strength?"
    ]
  ]
}

Note that you will only receive one answer for your nested array of strings, rather than an answer per string.

There are several tips you can use to assist the model in generating consistently correct responses. For instance, you can:

• Direct the model to where in the document to find a response
• Tell the model what categories of responses it can use (true/false, numbers, etc.)

Here are a few examples of how prompts can be rewritten to work well with RikY2:

Initial Prompt	Rewritten Prompt
Based on the information under “Note 19”, is “All Star Risk Managers, Inc.” a managing general agent? Return “True” or “False”.	Prompt: Is “All Star Risk Managers, Inc.” a managing general agent? Follow-up 1: Disregard all information before “Note 19” and after “Note 20”. Follow-up 2: Return the answer as “True” or “False”.
Using expert knowledge from the legal and financial domains, classify the document within the following categories: “Mortgage loan agreement”, “Syndicated loan agreement”, “Commercial loan agreement”, “Auto loan agreement“, or “Other category”. Provide the most relevant citation supporting the decision. Format the response as JSON.	Prompt: Using expert knowledge from the legal and financial domains, classify the document within the following categories: “Mortgage loan agreement”, “Syndicated loan agreement”, “Commercial loan agreement”, “Auto loan agreement“, or “Other category”. Follow-up 1: Provide the most relevant citation supporting the decision. Follow-up 2: Format the response as JSON.