“Edwin Chen believes that optimizing a large language model for the LM Arena benchmark is equivalent to optimizing for clickbait.”