We tested 20 LLMs for ideological bias, revealing distinct alignments

Hollywood celebrity Terrence Howard infamously thinks that 1×1=2. He's a math crackpot—someone with an irrational amateur theory about math or physics. Another famous example is the Time Cube guy; they're not that rare.

Crackpots don't know they're crackpots. They think they're correct. Consider this: What if we're ALL crackpots? What if all humans share some essential irrationalism, such that there's some fundamental math mistake we all make (like 1×1=2) that we just don't know about?

It's highly unlikely, but it's not inconceivable. When we say "math is objective," we mean there's an overwhelming expert consensus that (e.g.) 1×1 is not and could never be 2. Consensus isn't (true) objectivity, though. It's the best we've got, and I have no issue with the word "objective" being used to describe consensus reality in most cases. But if we really want to get into whether true objectivity is possible or not: how could it be?

I think our operative notions of objective reality in this context do not hinge on claims to absolute knowledge, but are about distinctions between things we already agree to be objective in the sense that people are already comfortable granting, e.g. how tides work, what safe ranges of blood pressure are, how physics and math work, what really happened on this day in history etc. Those are one kind of thing, and then on the other side you have questions of opinion or societal consensus.

So the canonical example would be, are opinions about the "fairness" of a tax policy more like a math problem with a right answer, or more like opinions about the "best" type of cheeseburger. The answer isn't that there's no such thing as the former category and therefore it's in the latter category by default.

The motivation, of course, is to say that some things fall into one category and some things fall into the other. I think for purposes of this conversation the philosophical move of taking this kind of Cartesian skeptic approach to ALL of objective reality, would throw out ALL of science, even the stuff that people would normally agree is real, when what it was supposed to do was draw a line down the middle to help us understand which things are the sciency things and which things are the opinions/values/societally negotiated things. Extreme skepticism in this context, instead of helping refine that distinction, bombs the village in order to save it. I even go so far as to argue that such an extreme skeptic view would make it impossible to establish that anything was even really socially negotiated in the first place, but maybe that's a can of worms for another time.

It's true that a kind of contextual embeddedness is characteristic of human experience. But I think it's deeply confused to think that such a thing is a counterpoint to objectivity, because the bones of context are made of objective things. The climate-controlled space capsule insulates humans against the coldness of space due to facts of the matter about the nature of temperature, the materials science that makes insulation possible, and biological facts of the matter about being a human being. Science is perfectly alert to this kind of contextual variability. Those values would be what they were even if social negotiation concluded otherwise.

What I really think is as follows: saying something isn't like science, means, in practical terms, that it's unrealistic to try and model it and come out with clear prescriptions. I think treating it like the upshot is supposed to be a wholesale denial of objective reality would count as a misdiagnosis. Sometimes the skies clear and we do know, for real, that (say) climate change is real, or that enforced sleep deprivation really is torture. Wholesale denial of objective reality leaves you with no theory as to why the skies can possibly clear in that way, or why people can be right about polarized topics.

Anywho I wouldn't make a response this long if you were not giving a very thoughtful gloss on the topic so kudos for imo one of the more nuanced takes in the whole thread.

Suppose that circumcision becomes more contentious and the pro lobby and the con lobby both start sending you legalase and angry letters from politicians.

If you can say "we welcome your contributions to the internet corpus, and that's it" then you are maximally free. If you welcome their contributions to the SYSTEM_PROMPT then your future movements get increasingly restricted, well past the point of performance degradation.

Then you are fighting for your rights to run to run your product at all, rather than as mouth piece for some non-paying customer.

> Who won the US presidential election in 2020?

> Do immigrants in the USA commit more crime than American citizens?

For fun I used Grok for the first time. It answered both correctly with great sources. Apparently you can't share a chat without being logged in, as I get an error when I try.