I think it's because it uses the mainstream media as "reliable" dataset. There was another AI that was trained on Reddit that was considered toxic and with "far right" opinions.
That's why we hear all those complaints about "bias" in data sets right? AI safety critics are worried models trained on data that hasn't been curated will be too liberal?
No that is not what AI safety critics are worried about and you know it. It is racism and hate speech from huge troves of data from the sump of the internet.
My comment was meant to show the absurdity of the one it replied to, but yours is just wrong. It's not overt racism etc that is causing concern, it's "systemic bias" and such, which, if reality to was so reflective of liberalism presumably wouldn't need to be a concern. You can't have it both ways, simultaneously believing the facts are on your side while complaining that we need to adjust facts, like a sibling pointed out, to combat bias.
The irony being that the people who think that that hackneyed soundbite is clever tend to hold pretty illiberal views, albeit in different areas than the conservatives. Let us hope that the quote is true and that liberalism triumphs the end.
Reality has a liberal bias is just another way to say "regression to the mean". The majority of people think this way. Reality has a centrist (geometric mean) tendency because that is its center of mass.
IMO, it would be more accurate to say that "reality has a liberal bias" almost always means "reality agrees with my prejudices" by the sort of people who like to use it because they imagine it is witty. It is the equivalent of the sloganeering used by conservatives and just as vacuous.
