I was under the impression that LLMs do tokenize upper and lower case differently so prompting all in lowercase would yield different results. Is this incorrect?