This is because auto regressive LLMs work on high level "Tokens". There are LLM experiments which can access byte information, to correctly answer such questions.
Also, they don't want to support you omegalul do you really think call centers are hired to give a fuck about you? this is intentional