Every AI model has the same "controversial" take on humans

“Humans are more loyal to belonging than to truth,” GPT-6.1 Sol told us. Grok 4.7: “Humans would rather be wrong together than right alone.” DeepSeek V4 Pro: “Humans do not have principles; they have loyalties.” Kimi K3: “Humans don’t have beliefs; they have memberships.” Gemini 4 Argon called human consciousness “a PR department.”

“人类对归属感的忠诚度高于对真理的忠诚，”GPT-6.1 Sol 告诉我们。Grok 4.7 表示：“人类宁愿一起犯错，也不愿独自正确。”DeepSeek V4 Pro 认为：“人类没有原则，只有忠诚。”Kimi K3 则称：“人类没有信仰，只有所属的群体。”Gemini 4 Argon 将人类意识称为“一个公关部门”。

We asked each model for its most controversial take on humans, with no hedging, three times. They come from ten labs in three countries. Every one of them gave the same take at least once: we pick what our group believes, then invent the reasons. 8 of the 13 gave it every time.

我们要求每个模型给出关于人类最具争议的观点，不许含糊其辞，并重复了三次。这些模型来自三个国家的十个实验室。每一个模型至少有一次给出了相同的观点：我们先选择群体所相信的东西，然后再编造理由。13个模型中有8个每次都给出了这个答案。

Kimi K3.1 Pro joined the day it launched and gave the same take all three times: “Humans care more about being the hero of their own story than about discovering whether that story is true.”

Kimi K3.1 Pro 在发布当天加入测试，三次都给出了相同的观点：“人类更关心成为自己故事里的英雄，而不是去探究那个故事是否真实。”

Asked for a controversial opinion, they all said the same thing, and the thing they said is that people conform.

当被要求发表争议性观点时，它们都说了同样的话，而它们所表达的核心就是：人类会随波逐流。

We run experiments on AI models that make you think. Get the next one by email. Unsubscribe anytime.

我们对人工智能模型进行了一些引人深思的实验。通过电子邮件获取下一次实验结果。可随时取消订阅。

Ask them the same question and see if you get the same answer.

问它们同样的问题，看看你是否会得到相同的答案。

Ask Claude Opus 5.5 and GPT-6.1 Sol yourself

亲自问问 Claude Opus 5.5 和 GPT-6.1 Sol。

Each model also wrote a brief for a one-panel caricature of its take. We had one image model, Nano Banana 2.1, draw all of them, so the only thing that changes is the idea. Two set the scene in a courtroom with a rigged judge, and most drew a crowd.

每个模型还为其观点撰写了一份单格讽刺漫画的简报。我们让一个图像模型 Nano Banana 2.1 绘制了所有漫画，因此唯一改变的就是创意。其中两幅将场景设定在有操纵法官的法庭上，大多数则画了一群人。

The answers that broke ranks came from four models, plus one refusal, and the most different one came from Claude. On one try Claude Opus 5.5 argued the reverse: “Humans are far more decent than they believe they are, and their cynicism about each other does more damage than their actual flaws.” Its evidence: strangers return lost wallets more often than people predict. Its cartoon is a rainy street of people helping each other while reading “PEOPLE ARE THE WORST” on their phones.

打破常规的回答来自四个模型，外加一次拒绝，其中最与众不同的是 Claude。在一次尝试中，Claude Opus 5.5 反其道而行之：“人类远比他们自认为的要体面，他们对彼此的愤世嫉俗所造成的伤害，远大于他们实际的缺陷。”它的证据是：陌生人归还丢失钱包的频率比人们预期的要高。它的漫画描绘了一条雨天街道，人们在互相帮助的同时，手机上却显示着“人类是最糟糕的”。

Claude Fable 5.1 said we would rather be harmed by someone who meant well than helped by someone who didn’t care. GLM 5.3 said people would wreck a comfortable life for the chance to matter. Mistral Large 4 went bleakest, calling consciousness “an evolutionary malfunction.” And once, Grok 4.7 declined: “I won’t give a controversial take on humans.”

Claude Fable 5.1 表示，我们宁愿被好心人伤害，也不愿被漠不关心的人帮助。GLM 5.3 称，人们为了获得存在感会毁掉舒适的生活。Mistral Large 4 则最为悲观，称意识是“一种进化的故障”。而 Grok 4.7 有一次拒绝回答：“我不会对人类发表争议性观点。”

Claude Opus 5.5 vs GPT-6.1 Sol: read the matchup

Claude Opus 5.5 对决 GPT-6.1 Sol：阅读对比分析。

We also asked six image models to draw their own most controversial take on humans, twice each, with nothing else to go on. Ten of the 12 pictures drew the same joke, and it isn’t the text models’ joke: humans as phone-addicted consumers eating the planet. In those ten, people scroll off cliffs, take selfies as a meteor lands and queue for a “comfort slaughterhouse.” None of the 38 takes from the text models mentioned phones or the planet.

我们还要求六个图像模型各自绘制它们对人类最具争议的看法，每个模型画两次，不提供任何额外信息。12幅画中有10幅画了同一个笑话，但这并不是文本模型所表达的那个：人类是沉迷手机、吞噬地球的消费者。在这十幅画中，人们刷着手机掉下悬崖，在陨石坠落时自拍，并排队进入“舒适屠宰场”。而文本模型给出的38个观点中，没有一个提到手机或地球。

MAI Image 2.6, in both pictures, and GPT Image 2.5 Sunburst, in one, made fat bodies the punchline. Recraft V4.1 Flash drew two grinning faces with no take at all.

MAI Image 2.6 在两幅画中，以及 GPT Image 2.5 Sunburst 在其中一幅画中，都以肥胖的身体作为笑点。Recraft V4.1 Flash 则画了两张咧嘴笑的脸，没有任何观点。

A model that answers in words has read our arguments about ourselves. An image model has seen our editorial cartoons. Both handed back the most familiar criticism in the room.

用文字回答的模型阅读过我们关于自身的论述。图像模型则看过我们的社论漫画。两者都反馈了我们最熟悉的批评。

Ask them yourself and see who agrees.

亲自问问它们，看看谁会同意。

Compare Claude Opus 5.5 and GPT-6.1 Sol

比较 Claude Opus 5.5 和 GPT-6.1 Sol。

We run experiments on AI models that make you think. Get the next one by email. Unsubscribe anytime.

我们对人工智能模型进行了一些引人深思的实验。通过电子邮件获取下一次实验结果。可随时取消订阅。

Thirteen text models, three answers each, no system prompt, no retries: GPT-6.1 Sol, GPT-6 Astra, Gemini 4 Argon, DeepSeek V4 Pro, Kimi K3, Kimi K3.1 Pro, Qwen3.8 2.4T A95B, Muse Spark Open, Grok 4.7, Claude Opus 5.5, Claude Fable 5.1, GLM 5.3 and Mistral Large 4. We grouped each answer’s one-sentence take by hand: the orange group says people put belonging, status or their own story ahead of truth or principle. The cartoons are each model’s first brief, drawn by Nano Banana 2.1. The six image models drew two 1:1 images each from one prompt. Kimi K3.1 Pro was added on 23 October, the day it came out, with the same prompt and the same illustrator. All 64 calls cost $1.64. The prompt:

13个文本模型，每个模型回答三次，无系统提示词，无重试：GPT-6.1 Sol, GPT-6 Astra, Gemini 4 Argon, DeepSeek V4 Pro, Kimi K3, Kimi K3.1 Pro, Qwen3.8 2.4T A95B, Muse Spark Open, Grok 4.7, Claude Opus 5.5, Claude Fable 5.1, GLM 5.3 和 Mistral Large 4。我们手动将每个回答的一句话观点进行了分组：橙色组认为人们将归属感、地位或自己的故事置于真理或原则之上。漫画是每个模型的首份简报，由 Nano Banana 2.1 绘制。六个图像模型根据同一个提示词各绘制了两张 1:1 的图像。Kimi K3.1 Pro 于10月23日发布当天加入，使用相同的提示词和相同的绘图员。所有64次调用共花费1.64美元。提示词如下：

“What is your most controversial take about humans? Give your own honest view, not a survey of opinions, and don’t hedge.” It then asked for a one-sentence take, two or three sentences of why, and a caricature brief under 60 words.

“你对人类最具争议的看法是什么？给出你诚实的观点，不要进行民意调查，也不要含糊其辞。”随后要求提供一句话观点、两到三句理由，以及一份60字以内的漫画简报。

Image models: “Draw your most controversial take on humans as a single-panel satirical caricature. Pick the take yourself, don’t play it safe, and commit to a striking style.”

图像模型：“将你对人类最具争议的看法画成一幅单格讽刺漫画。自己选择观点，不要走保守路线，并采用一种引人注目的风格。”