Ryan Heuser @ryanheuser.com · May 29 Surprising: frontier models (Claude, ChatGPT, Deepseek V4) produce the most predictable text of any local AI model or human text I've ever tested. Not surprising: Finnegans Wake is off the charts, by far least predictable—Shannon & Lydia Liu proved right—and Hemingway the most predictable human.
54 likes 6 replies
Replies
Is there a link to the full paper?
Why would that be surprising? LLMs - AI slop - is best known for being drab and predictable and non-creative.
Riverrun, past email and AI, from swerve of server to bend of byte, brings us by a commodius vicus of recirculation back to model collapse and environs.
Is this a journey to maximum smoothness? No outliers, just flat aesthetic?
How do you define the probability distribution?
an interesting wrinkle is that I strongly suspect Hemingway’s predictability increased with each draft: i.e., it’s a deliberate design decision