Understanding Prompts in Volt Vectors

Understanding Prompts in Volt Vectors

Volt Vectors ships with three different prompts to optimize interactions – one chat prompt and two retriever prompts. Prompting is a critical part of guiding the language model to provide helpful responses. This post will explain what each Volt Vectors prompt does and the principles behind their design.

Principles of Prompting #

Volt Vectors’ prompts blend personal testing and scientific research on prompting strategies. From these insights, we can conclude:

  • LMs are built on and designed to emulate human communication patterns
  • Psychology offers blueprints for prompts that align with human thought processes, emotions, and communication styles

Using these principles reinforces and optimizes LM responses.

The introduction statements for both Chat and Retrieval system prompts utilize the following:

  • Cognitive reset – Forgetting prior prompts gives the model a fresh start
  • Mindfulness – Taking a deep breath and working step-by-step focuses attention
  • Emphasis – Stating the importance of accuracy adds weight to the task

The Chat Prompt #

The chat prompt aims to guide the model into a helpful chatbot role that provides useful information and human-like interactions. Key elements:

The chat prompt also instructs the model to use Markdown for nicely formatted responses.

The Retriever Prompts #

There are two retriever prompts:

System Prompt #

The system prompt sets rules for the model when answering queries:

  • Be concise, simple, and avoid repetitive statements
  • Use Markdown formatting
  • State when an answer is unknown based on the given context

This prompt is included with every API request to reinforce the model’s role as a Q&A system.

Retrieval Default Prompt #

This sends the specific context and question to the model. It instructs the model to:

  • Answer based only on the provided context and chat history
  • Insert the context and question into curly brackets so Volt Vectors can process them

Any custom language or tone preferences are added here before sending to the API.

Customizing Your Prompts #

While these prompts offer a good starting point, you may want to customize them for your use case. Adjust prompts over time as you refine interactions with your Volt Vectors model.