I thought SOTA was still a few % in difference? It seems worrisome that they few percent might be making a coin flip right-randomly instead of wrong-randomly o

I thought SOTA was still a few % in difference? It seems worrisome that they few percent might be making a coin flip right-randomly instead of wrong-randomly on a mislabelled subgroup of data...