6 ms·

Very telling of what?

• nsingh2 10mo ago
  Telling of where the boundary of competence is for these models. And to show that these models aren't doing what most expect them to be doing, i.e. not counting legs, and maybe instead inferring information based on the overall image (dogs usually have 4 legs) to the detriment of find grained or out-of-distribution tasks.