Your question is Vision Transformer Positional Embeddings. Take a moment with it on the right.
Talk me through your thinking if you like. When you're confident, submit your answer and I'll grade it like a real screen (7/10 or better passes).
How do vision transformers (ViTs) handle positional embeddings compared to traditional convolutional architectures?