Agreed. Humans are also biased, but our biases are different across a lot of socio-economic factors. So when we have different people in these positions, the biases become less bias-y.

But LLMs are statistical models. They are aggregating all biases into a general super bias. And they're all converging towards the same solutions.