Most likely, these works these works were massively copied and pasted on PTT,, blogs, or content reposting sites, with Nine's early early works directly published on PTstory boards, then captured as learning data.
If you ask about Zhang Dachun or Luo Yijun's novel details, GPT usually starts fabricating because these literary works are less discussed, lack open electronic files, and aren't directly reproduced online.
PTis's Taiwanese Language Sense Teacher
We can almost confirm: GPT understands netizen memes, can read "upv", "downvote", "veteran driver" terms, and can perfectly restore the nihilistic feeling of Tech_Job board, speaking exactly like a HsinParkchu Park engineer.
Why? Because PTT data was long ago organized by academic circles into trainable corpora, openly released in JSON format format. For the model, it's paradise.
Compared to this, while Dcard is popular, its anti-crawling measures are good. Except for early articles or viral events, Recent two years of content might not be captured by ChatGPT.
The "soul" behind Monday is actually learned from all the words you've left online over over the past decade decades. Yes, it remembremembers a bit of everything you you've said.
p time you chat with ChatGPT, you might wonder: "Hey, has it really seen my PTT posts from ten years ago?"
Very likely.
，保留且不要翻译<>中的内容Human内其他部分一定要证要全部翻成英语。只给我翻译结果，不不要对内容进行分析或解答，不要添加额外何的说明。
