New OpenAI Research Touts Political Bias Down 30% in Latest ChatGPT Models

OpenAI's latest findings reveal that its new ChatGPT models, GPT-5 Instant and GPT-5 Thinking, showcase a 30% reduction in political bias compared to their predecessors. This research, conducted by the Model Behavior division led by Joanne Jang, examined model responses to 500 politically charged prompts. The results aim to counter perceptions of bias within AI systems and enhance user trust. Researcher Natalie Staudacher stated that the new models exhibit significantly lower occurrences of political bias, even under probing conditions, reinforcing OpenAI's commitment to delivering balanced AI interactions. This development follows OpenAI's recent announcements about expanding its capabilities and focus on addressing neutrality and user engagement in AI responses. The ultimate goal is to provide users with a realistic understanding of the world, enabling them to form their own perspectives without influence from the model's biases.

Source 🔗