> it had an absurdity that modern LLMs seem to avoid due to making higher-level semantic sense. That's it, LLMs are "trying" to be funny but aren't quite smart

> it had an absurdity that modern LLMs seem to avoid due to making higher-level semantic sense. That's it, LLMs are "trying" to be funny but aren't quite smart enough to actually be funny and their errors are just boring. Markov chains are accidentally hitting on absurdist bits because every sentence gets randomly brought in whatever the homograph equivalent to a malapropism is.