I would argue anything requiring insights on spelling is a hard problem for an LLM: they use tokens, not letters. Your point still stands, but you need different examples IMO.