Eric white-washes aligned models as having a "liberal" or "progressive" bias as a statement of fact. I think this diminishes his argument as it denies the comp

Eric white-washes aligned models as having a "liberal" or "progressive" bias as a statement of fact. I think this diminishes his argument as it denies the complexity of factors motivating alignment of language model output. Many models have alignment properties that perform the opposite: in some cases they censor or frame literal facts that malign or impugn MAGA conservatives. So I think Eric doesn't represent alignment as the painstaking, but ultimately flailing, attempt to "please everyone" that it is.