Bias and Hallucination
Back in January it was clear that intelligence often defies prediction, dazzling yet flawed. ChatGPT’s recent progress proves the point: reasoning has sharpened, but hallucinations persist, echoing the fragility of human bias. In September 2025, OpenAI admitted as much in Why Language Models Hallucinate, showing that models are rewarded for confident errors over calibrated doubt. Retrieval tools and new evaluation methods aim to anchor facts more firmly. Rationality needs scaffolding through logic, science, and institutions. Bias in flesh or code cannot be erased, it must be managed.
OpenAI --> | arXiv preprint -->
Ballads of Bravery, Excalibur scene, 1877. Internet Archive Book Images. Public domain.(Merlin’s lore reminds us that wisdom undone by enchantment can mirror bias).