AI models describe humans as a bunch of amoral sheep

AI models have criticized humans, describing us as irrational, dangerous, and more devoted to belonging than truth. According to 12 different large language models, we are like lemmings who prefer group consensus over individual logic. GPT-6.1 Sol and Kimi K3, for instance, argued that we are more concerned with maintaining social connections than pursuing truth, while Grok suggested that our moral language is primarily used to judge others and justify our allies.

Not all models shared the same negative perspective, with Claude Opus 5.5 believing that despite our flaws, humans are actually decent and that our cynicism does more harm than our genuine shortcomings. Mistral Large 4, however, painted a more bleak picture, describing us as "terrified animals wearing the costume of rationality" and constructing elaborate myths to avoid acknowledging our temporary and meaningless existence.

When image models were asked to provide their most controversial take on humanity, most of them depicted us as phone-addicted consumers who harm the environment, similar to editorial cartoons that have been widely shared over the past two decades. ModelCompare's developer, Rakan Nimer, found the results surprising but not unexpected, as these AI models have learned from the same human arguments about ourselves.

He emphasized that the agreement among the models may not necessarily represent a genuine consensus but rather a reflection of our shared pessimism and should be approached with skepticism.