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

What happened

A news team asked 12 AI models from ten labs for a bold, no-hedging opinion about humans. They asked three times each. Almost every model gave the same basic take at least once. The take is that humans care more about fitting in with their group than about finding the truth. Seven of the 12 models gave this take every time.

Key numbers

• 12 AI models were tested.
• They come from ten labs in three countries.
• Each model was asked 3 times, so there were 36 answers in total.
• Every model gave the shared take at least once.
• 7 of the 12 models gave it all 3 times.

What the models said

The answers used different words, but the idea was the same. One model said humans are more loyal to belonging than to truth. Another said humans do not have principles, only loyalties. Another said human consciousness is like a PR department that makes up good-sounding reasons for instincts. The models were asked for a controversial opinion, yet they all chose the same idea about people.

Why it matters

The shared answer is interesting because it may say more about AI than about people. Many AI models learn from huge amounts of human writing. A common theme in that writing could make a model repeat it when asked for a bold view. Because the models gave nearly the same answer, the result shows how similar these systems can be, even when they come from different companies.