Surprising to see these models stumbling on what at first glance seems like a simple task, it would be interesting to see how the non-vision models fare if you

Surprising to see these models stumbling on what at first glance seems like a simple task, it would be interesting to see how the non-vision models fare if you convert the problems to ascii art