Update model-card-annotated.md (#1418)
Proposing a rephrasing to move away from the binary "can the model to X" to "where risks pertaining to X evaluated"
For example for NCII, most image generation models that can generate some level of partial nudity can generate images of some quality that may qualify, but concerns of misuse will be drastically different depending on how easy it is to achieve those results, how realistic they look, etc.
