From Domain Segmentation to Reading Domain Signals
Can a general-purpose open model assess brand threat without any task-specific training? We tested four model configurations on 25 public domain disputes and found that they already recognize many of the right signals. But their readings change when you change which facts are presented first.
Read →