Study finds AI models converge on same controversial view of human nature

Researchers conducted an experiment asking 12 different AI models to share their most controversial opinions about humans, repeating the prompt three times each. Across 36 total responses, 29 gave essentially the same answer: that humans prioritize social belonging and acceptance over truth-seeking. The consistency across different models and multiple runs suggests the models converge on this particular critique despite being built by different organizations with different underlying approaches.

The finding raises questions about whether this represents genuine insight from the models, learned patterns from training data, or an artifact of how language models are built and aligned. It also highlights how current large language models tend to produce similar outputs when asked open-ended philosophical questions, even when developers have tried to differentiate them.