I was cosplaying GPT on the first joke about a pun. I read the text without ever pronouncing it and parsing it right into abstract ideas on a literal level. It

I was cosplaying GPT on the first joke about a pun. I read the text without ever pronouncing it and parsing it right into abstract ideas on a literal level. It took me 30 seconds and reading it aloud to understand it. Had anyone told me this joke aloud, my reaction time would be much shorter. This joke relies on a level of senses and context that language model does not have. Ironically, when GPT-5 or 6 is out, this very post is going to be included into the corpus, LLM will learn to discard its own previous irrelevant explanations and infer the explanation of the pun. It is all fun and games to find examples like that but the rate of progress in this field is nothing short of staggering.