Show Input Space

Purpose:

• Dimensionality Reduction Visualization: View a reduced-dimensional plot of your LLM interactions, where each dot represents a single API call.

• Cluster Analysis: Identify clusters of similar input data and see how different prompt families perform across these clusters.

• Identify Edge Cases: Spot regions of poor performance (often color-coded) to pinpoint specific areas where prompts need refinement or additional testing.

Example: Consider an app that processes various types of web-scraped content. By using the Show Input Space tool, you can see clusters representing different content types (e.g., news articles, blog posts, research papers). If you notice that a cluster corresponding to research papers is predominantly blue (indicating lower accuracy), you know that you should focus on developing a more tailored prompt family for that content.