The potential and observed use of language models and generative AI systems to create false news articles, misleading headlines, or fabricated narratives at scale. This poses significant risks to information integrity and public trust.
Overview
As language models improve in fluency and coherence, concerns grow about their use for generating convincing fake news without explicit human authorship. Early work on GPT-2 demonstrated the capability to generate coherent news-like text; later models show further improvements, raising questions about detection, attribution, and responsible deployment.
Key papers
- 2019-solaiman-gpt2-release-strategies — early analysis of GPT-2's capacity to generate fake news and strategies for responsible release