Perhaps it is part of the training where humans are involved. Maybe the trainers were frustrated English Lit majors. That could account for several deficiencies in the LLMs.  :)

I think that's true to some extent, and how they're explained to work at a very high level, but it's not purely a statistical thing. If it were, we would've had LLMs a long time ago.

k00b

Stacker Saloon

saloon

The only reason you see this a lot in LLMs is that their training included a lot of it. The LLMs are just big statical, this word comes after that word, language generators. To distinguish yourself from an LLM, you just have to do things that are not very statistically likely.