multimodal — A model that works with more than text — reading images, audio, or video, and sometimes generating them too.
Why it matters
A working taxonomy for picking a red-team method by threat model, and a path for turning one-off qualitative findings into repeatable automated evaluations.
