max_length: 256
(Whisper: This also matched my gut feeling—LLMs have subtle word choice patterns, and even a Naive Bayes classifier should catch them. I just didn’t expect the signal to be this strong.)
。业内人士推荐体育直播作为进阶阅读
ВсеПолитикаОбществоПроисшествияКонфликтыПреступность，详情可参考同城约会
Фото: Aurore Marechal / Getty Images
Next, I used an LLM to generate chapter summaries of these texts, then fed the summaries back into various LLMs to regenerate full articles. This gave me a similar-sized, diverse dataset of LLM-generated texts closely aligned in topic and style with the human-written ones.