Does well in my tests, limited as they were, but it did well in zero-shot tasks in niche domains historically (and possibly still here) underrepresented in trai

Does well in my tests, limited as they were, but it did well in zero-shot tasks in niche domains historically (and possibly still here) underrepresented in training data (microscopy)