LLMs and other ai models are applied statistics, and the complete average of all information online is uh. More than a bit right-leaning due to the sheer volume of bigoted screeds, even without getting into the fact a lot of the companies that decided what are 'good outputs' were mewling centrists at best, yeah.
The idea of an unbiased AI model is not real when almost all text and images it was trained on will have implicit messaging, even if the model cannot 'think'. 