Few-shot Researcher-In-The-Loop Screening/Annotation

[adelavega]

Kind of like ASReview, or DSPy-- Use human-in-the-loop, example-driven prompt development.

That is given a few positive and negative examples of selection/annotation, improve the prompt (either generatively or through few-shot examples) to improve performance vs one-shot prompting

Evaluation: Can use neurometabench to feed in examples in a cross-validated manner