Combined with OCR it can be trained to review scans and pdfs of forms and diagrams and mine pertinent data from them, cutting down on manual time to data mine with enough accuracy you can halve (or more) the required time with data checks to generate good data. The LLM stuff is also pretty handy for summarizing large bodies of text and spitting out specific data, training modules, even podcasts discussing the contents.

I like it for initial note taking and comparison analytics.