How to measure LLM writing quality when there is no right answer?

Try Voice Writer - speak your thoughts and let AI handle the grammar: https://voicewriter.io How do you evaluate the writing quality of LLMs which is inherently subjective? Unlike multiple-choice benchmarks like MMLU where answers are clearly right or wrong, this is not possible for natural language generation. There are a few approaches. When a reference is available, metrics like BLEU, ROUGE, or BERTScore are useful, but they don’t fully capture fluency, tone, or coherence. Human ratings are the gold standard but come with their own biases. Finally, LLMs can also judging other LLM outputs w…