GPT-2: 1.5B release

OpenAI says GPT-2’s 1.5B model can generate highly convincing synthetic text that people often rate as credible.

Partner research found the model can be fine-tuned to produce extremist propaganda, including white supremacist, Marxist, jihadist, and anarchist content.

OpenAI released a detector that works well in tests, but it is not reliable enough to use on its own.

The findings underscore the growing realism of large language models and the dual-use risk of misuse, making detection a pressing challenge.