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

520 points · 513 comments on HN · read original →

Points and comments are a snapshot, not live.

12 AI models unanimously say humans prioritize belonging over truth.

ModelCompare asked 12 text AI models for their most controversial take on humans, three times each. 29 of 35 responses converged on the same idea: humans put group loyalty, status, or belonging ahead of truth or principle. Examples: "Humans are more loyal to belonging than to truth" (GPT-6.1 Sol), "Humans would rather be wrong together than right alone" (Grok 4.7), "Humans don't have beliefs; they have memberships" (Kimi K3). Seven models gave this take all three times. Four models broke ranks occasionally, e.g., Claude Opus 5.5 argued humans are more decent than they believe. Image models offered different critiques: humans as phone-addicted consumers. The study included 12 text models and 6 image models, costing $1.54 total.

What commenters are saying

Many commenters dismissed the finding as unsurprising: the models trained on similar data (Common Crawl) simply reflect a common human self-critique. One commenter noted the real interest lies in whether this convergence reveals something about human communication patterns or about near-identical training sets across labs. Some defended the claim's accuracy, citing evolutionary social bonding as our species' strength and weakness. A few pointed to the backfire effect-that facts often strengthen false beliefs-queuing a reply noting a 2020 review found widespread replication failures of backfire effects. One comment thread wryly suggested the models got the opinion from 4chan.