I suspect it’s because LLMs are smarter when they are verbose. The more they water down the model they serve the more they have to dial up the verbosity to comp

I suspect it’s because LLMs are smarter when they are verbose. The more they water down the model they serve the more they have to dial up the verbosity to compensate.