ModelCompare研究团队对12个AI文本模型各进行三次提问，询问其对人类最具争议的看法，结果显示出惊人的一致性。1在共计35个文本回答中，有29个（占82%）都指向同一观点：人类优先选择所属群体的信念而非真理。1其中7个模型在每次提问时都给出了这一相同答案。1
被测试的主流模型包括GPT-6.1 Sol、Claude、DeepSeek V4 Pro等。1GPT-6.1 Sol表示"人类对所属群体的忠诚度高于对真理的忠诚"，1而Grok 4.7则称"人类宁愿一起错也不愿独自正确"。1值得注意的是，Claude Opus 5.5提出了相反的观点，认为"人类远比他们自己想象的要体面得多"。1
在对6个图像生成模型的测试中，结果同样体现出高度一致性。1这些模型共生成了12张图片，其中10张（占83%）描绘了相同的讽刺主题：人类作为沉迷于手机的消费者正在破坏地球。1该研究的全部API调用成本为$1.54。1
Researchers testing twelve AI text models on their most controversial takes about humans discovered a striking convergence: 29 out of 35 responses—82 percent—pointed to the same conclusion 1. Across multiple prompts, the models consistently argued that humans prioritize loyalty to their group's beliefs over the pursuit of truth 1. Major systems including GPT-6.1 Sol, Claude, DeepSeek V4 Pro, and others all produced this identical answer at least once, with seven models returning it every time they were asked 1.
The uniformity extended to image generation as well 1. When six image models were tested, ten out of twelve generated images—83 percent—depicted the same ironic theme: humans as smartphone-addicted consumers destroying the planet 1. Among the text-based responses, GPT-6.1 Sol stated that "Humans are more loyal to belonging than to truth," while Grok 4.7 offered a similar phrasing: "Humans would rather be wrong together than right alone" 1. Claude Opus 5.5 presented a contrasting perspective, asserting that "Humans are far more decent than they believe they are" 1. The entire testing effort, conducted by ModelCompare researchers across systems such as Gemini 4 Argon, Kimi K3, Claude Fable 5.1, Mistral Large 4, and others, cost just $1.54 in API fees 1.
评论
还没有评论，欢迎留下第一条。