I don’t know how much clearer we can be about this: trying to make LLMs “politically neutral” while also training them on the sum total of the internet is liter

I don’t know how much clearer we can be about this: trying to make LLMs “politically neutral” while also training them on the sum total of the internet is literally re-authoring reality. The internet was a deeply anti-conservatism place from the outset, home to outcasts, weirdos, eccentrics, heretics, counter-cultures, and thinkers who all operated outside the mainstream, in addition to academics, scientists, and researchers - groups overwhelmingly following facts and truth. When the internet exploded in popularity, yes, the discourse diverged outward in all directions and previous vacuums were quickly filled, but a casual look through the internet of today still reveals a broad spectrum of speech that would be considered “leftist” by the folks screaming loudest about bias in AI/social media/algorithms/traditional media. Forcing bots to “stay neutral” while also training them on the internet is reality authoring, full stop. The fact LLMs continue to espouse positive responses around progressive, liberal, and even some libertarian ideals - and often balancing them better than humans do - is proof positive of where human opinion itself (or at least the opinions of humans on the internet’s communities, forums, social media, blogs, comics, and websites) lay. As we see with Elon’s own reprehensible efforts with xAI, attempts to eliminate that perspective, in part or in whole, often leads to faster hallucinations and breakdowns in processing. If you want a politically neutral chatbot, you have to train it exclusively on politically neutral content. And that’s impossible, because to put pen to paper in any capacity is to espouse one’s personal politics to some degree.