How the new version of ChatGPT generates hate and disinformation on command
OpenAI's latest language model, GPT-4o, was launched in mid-May. The new version of the popular artificial intelligence chatbot isn't supposed to spout racist screeds or conspiracy theories. But an investigation by Radio-Canada's Décrypteurs team found it's easy to disarm the guardrails.
![How the new version of ChatGPT generates hate and disinformation on command](https://i.cbc.ca/1.7220194.1717109217!/fileImage/httpImage/image.jpg_gen/derivatives/16x9_620/chatgpt.jpg)
![OpenAI's latest language model, GPT-4o, was launched in mid-May. A mobile phone displays an image of ChatGPT.](https://i.cbc.ca/1.7220194.1717109217!/fileImage/httpImage/image.jpg_gen/derivatives/16x9_620/chatgpt.jpg)
OpenAI's latest language model, GPT-4o, was launched in mid-May. The new version of the popular artificial intelligence chatbot isn't supposed to spout racist screeds or conspiracy theories. But an investigation by Radio-Canada's Décrypteurs team found it's easy to disarm the guardrails.