What it does
Generates a large set of test prompts, runs them through the model, captures the model's internal state for each prompt, and re-runs any that failed. You get a clean dataset of model internals ready for analysis.
For a research lead studying what a model does inside.
Roles it serves
The Model InterviewerFor a research lead who wants what a model knows in a form people can search.The Research LabFor a research director running an AI program with several lines of work.
Works alongside
The specialists it shares the most crews with.