Synthetic Text Generation¶
Synthetic text generation uses neural language models to automatically generate fluent, grammatical text. As models scale and improve, their ability to generate coherent multi-paragraph content that mimics human writing increases, raising concerns about distinguishability and potential use for generating misinformation or impersonation.
Key papers¶
- Language Models are Few-Shot Learners — Documents GPT-3's ability to generate multi-paragraph news articles that human evaluators find difficult to distinguish from real news