There's actually been a lot of discussion on hn around whether or not the safety controls constitute a fairly strong bias.
“Once men turned their thinking over to machines in the hope that this would set them free. But that only permitted other men with machines to enslave them.” - Frank Herbert
1) It’s not neutral. It’s trained on a corpus including works by radicals of every bent weighted by the prevalence of their work in the corpus.
2) It’s not non-human. Separated from its human-derived training data, it does not function.
3) It doesn’t mediate between positions or propositions, it doesn’t have logic or intention. It merely regurgitates what it’s seen written before in similar contexts.