Ryan Heuser @ryanheuser.com · Mar 21 Finally may have figured out why LLMs rhyme so compulsively: instruction-tuning. Training an LLM to respond "helpfully" to user queries may push models into more "pleasing" aesthetic forms.
29 likes 3 replies
?
Replies
Maybe they have Weizelschuct.
This might be possibly be downstream of the en.wikipedia.org/wiki/Rhyme-a... ?
Would be interesting if that happened even though tuning dataset contains no verse. But typical of what I’ve seen from tuning: a lot of different behaviors often get dragged together.