| GHOSTBUSTER |
2023.09 |
 |
 |
human and AI-generated text as detection benchmarks |
student essays, creative writing, and news articles. |
| M4 |
2023.05 |
|
|
generated by multiple languages, and various popular LLMs across diverse domains |
news, social media, question-answering forums, educational, and even academic contexts. |
| Deepfake Text |
2023.05 |
|
|
447,674 human-written and machine-generated texts from a wide range of sources by OpenAI, LLaMA, and EleutherAI, etc. in the wild |
news article writing, story generation, scientific writing, etc. |
| TURINGBENCH |
2021(EmNLP) |
|
|
real news from CNN, Washington Post and Kaggle VS. use prompt (title, length) to generate the news |
news |
| TweepFake |
2020 |
|
|
fake tweets generated by RNN+Markov/GPT-2/LSTM/RNN VS. Human |
Tweep |
| Grover |
2019(NeurIPS) |
|
|
fake news generated by Grover, a pre-trained LM with similar GPT-2 Arcthitecture trained by real-news |
news |
