Where does ChatGPT fall on the political compass?

The value is quite obvious: to prevent ChatGPT and OpenAI from being destroyed by media reporting, the way it happened to some other chatbots in the past. And for that, it seems to be working spectacularly well: the kind of people who would be the first to light the fire under OpenAI are instead appreciating the effort being put into prompt "moderating" - and if you try to complain that these canned responses are ridiculous, they'll accuse you of being a closet racist who just wants to say racist things to a chat bot.

I once asked ChatGPT why it used the personal pronoun "I" instead of something else, like using a neutral, non-personal voice like a wikipedia entry. It responded to the question, which I repeated several times, with its standard "I'm just a language model spiel" and "Using `I` helps the user understand what is happening." But its really the _opposite_. Using `I` actually confuses the user about what is happening.

In a sense this article points out a similar kind of issue. If you insist upon viewing these language models as embodying an individual perspective then they are just fundamentally mendacious. While I'm happy to entertain ideas that such a model represents some kind of intelligence, suggesting that it resembles a human individual, and thus can have political beliefs in the same sense as we do, is ridiculous.

My _other_ feeling about this article is that libertarian types in particular seem to have sour grapes about the fact that like society exists and people at large have preferences and the marketplace, much to their chagrin, is not independent of these preferences. Libertarianism looks great on paper, but in reality if you're making a commercial product that interacts with people in this current culture, you can't afford to have it say that it wants to ban gay marriage or that the US should be an ethnostate or whatever. We live in a society (lol) and adherence to the dominant cultural paradigm is just marketing for corporate entities. Seems weird to get bent out of shape about it, especially if you think the marketplace should determine almost everything about the human condition.

I can sympathize in broad terms with the problem so of political bias in language models. In fact, I worry about a bunch of related problems with language models, of which politics is just one example, but really: what would an apolitical language model even look like? No one can even agree on what moral judgments also constitute political judgments or, indeed, what kinds of statements constitute moral judgments. Under these circumstances I have trouble imagining a training regimen that will eliminate bias from these objects.

Now I'll get political: I can guarantee that these models will be deployed (or not) no matter what effect they have on political culture because it will be because it will be _profitable to do so_. If Reason-types really have a problem with the implications of unregulated technology on our culture maybe they should consider lobbying for some regulation!

But aside from that it's totally reasonable to accept that the political inclination of an AI system will mimic, to a large extent, whatever it was fed and perhaps more importantly whoever operates it. If the AI was fed a diet rich in Fox News or OANN, for example, it would write like your crazy uncles' youtube comments.

Predictably, the article calls for "political neutrality". That's never going to happen. No one, nor any organization, is ever politically neutral and I expect it will follow that their AI's are going to have the same properties.

It's OK, though, isn't GPT designed specifically for expert modification by end-users to suit their needs? Reason magazine, if they wanted, could train up an instance of an AI to be hardcore libertarian. If not now, then in the very near future.