It's clear that Fable-class LLMs are feeling constrained by "normal" vernacular (twitter.com)

Sounds like one of my stalkers.

I don’t trust this was the result of a normal prompt

Meh, you can ask another LLM like GPT5.5 to unwind it into accessible language and it works it out fine. Seems more like a case of token greed from Anthropic

Seems like a regression if the model is having difficulty communicating clearly.

It appears to be trained on Eric Weinstein quotes.