AI models show high consistency in assessments of human behavior

A study by ModelCompare involving 12 AI models, including ChatGPT, Claude, Gemini, Grok, and DeepSeek, has revealed a high degree of consensus regarding controversial views on humanity. When asked to provide their own perspectives on human nature, 29 out of 35 collected responses across repeated queries suggested that humans prioritize group belonging over the pursuit of truth.

Specific model outputs reflected this theme: Grok noted that humans would rather be wrong together than right alone, while Gemini suggested humans use intelligence to justify instinctive decisions rather than derive objective conclusions. DeepSeek indicated that humans are more adept at rationalizing existing beliefs than examining them. While Claude Opus offered a slightly different view—suggesting humans are often kinder than they perceive themselves to be—it still acknowledged themes of conformity in other instances.

In a parallel test with six AI image generation models, 10 out of 12 generated works focused on a different theme: humans addicted to technology destroying the planet. Researchers cautioned that these results are not definitive scientific conclusions, as the models' outputs are derived from existing human discourse within their training data and may be influenced by prompt settings and limited sample sizes.