If the 90% of the internet says ai data centers are bad, statistically the ai is heavily going to repeat those sentences it was trained on.
Don't assume the billionaires who are in charge of training their AI won't use a first iteration of an LLM to curate/filter the training dataset of the next iteration of an LLM.
The problem is that an LLM is not an inherently fair or impartial tech. It will mimic whatever it was fed during the training phase. So whoever controls the training phase, controls the LLM's disposition.