I don’t know if this is possible but I think it would help avoid repetition when requesting creative writing from ChatGPT if it kept track of the last time it used the next token it plans to employ. I’ve noticed it uses for example the word ‘determination’ a lot, if it was trained that any average body of work uses only this word x number of times and if it used it once it reduced the likelihood of selecting it as the next token(though perhaps never discarding it entirely for paragraphs requiring its repeated use), keeping it within a certain range based on its presence in one or more texts used in the training. Likewise keeping track of how many number of words it takes between repeating any given word on average(‘determination’ should only appear every x number of words on average in a body of text x number of words long etc), some would score higher, others lower etc. This could be applied to number of adjectives, verbs in certain tenses, etc. Idioms, or groups of words. Or more abstract concepts such as how many characters die or are born in a text.
Just my two cents, I looked for similar subjects previously discussed but couldn’t find any. Still becoming familiar with the forum layout so apologies if I missed someone posting something similar before.
Thank you for reading.