Yeah, but it clearly breaks down the spelling correctly in it's reasoning, e.g. a letter per line. So it gets past the tokenization barrier, but still gets hopelessly confused.