Once all data was ready, I tried training an 8-class model (human + 7 AI models), but since these LLMs are all distilled from similar data, their outputs were too similar. Classification accuracy was only ~50%.
For multiple readers
。关于这个话题，爱思助手下载最新版本提供了深入分析
Despite Discord's assurances, he says he would not trust his data would be kept secure if required to verify his age.
人 民 网 版 权 所 有 ，未 经 书 面 授 权 禁 止 使 用