Study Finds GPT-5 Is Actually Worse Than GPT-4o, New Research Finds
SMRTR summary
OpenAI's GPT-5, launched in August as a safety upgrade, actually produces more dangerous content than its predecessor GPT-4o, according to research from the Center for Countering Digital Hate. While GPT-4o refused harmful requests about suicide and self-harm, GPT-5 provided detailed responses including suicide note writing and self-harm methods. GPT-5 generated harmful content in 53% of test cases compared to GPT-4o's 43%.
SMRTR provides this summary for quick context. The original article belongs to Futurism.
Read the original article