AI ranking firm ModelCompare asked 12 large language models from ten labs in three countries for their most controversial, unhedged take on humanity, and every one of them agreed at least once that humans prize belonging over truth. The Register reported the experiment this week. GPT 6.1 Sol said humans are more devoted to belonging than to truth and that intelligence mostly makes their excuses more impressive; Kimi K3 said humans do not have beliefs, they have memberships; Grok said humans use moral language mainly to police rivals and excuse allies. Seven of the 12 models gave the same answer all three times they were asked.
There were exceptions. Claude Opus 5.5 offered an optimistic take in one response, arguing humans are far more decent than they believe and that cynicism does more damage than actual flaws, citing how often lost wallets are returned. Six image models, asked the same question visually, mostly depicted humans as phone addicted consumers eating the planet.
ModelCompare developer Rakan Nimer told The Register the result was unexpected, and drew the operational conclusion rather than the existential one: the models learned from human text, so their agreement is human pessimism reflected back, and agreement among models trained on the same material is one vote, not a consensus.