What's the difference between a paper on a new prompt and a paper discussing a new domain-specific model, e.g. heart failure risk? If they analyze the problem and solution equally, they both seem useful. It's not like most other ML papers share their weights or datasets.