My task is in the NLP space, maybe that makes it more approacheable - information extraction from semistructured documents. I can do extraction from existing documents with GPT-3 (question answering) or I can generate new data with known tags.