Comment by landl0rd | Hacker News Reader

Comment by landl0rd | Hacker News Reader

Parent Full thread

Most frontier LLMs skew somewhere on the libleft quadrant of the political compass. This includes grok 4 btw. This is probably because American "respectable" media has a consistent bias in this direction. I don't really care about this with media. But media outlets are not "gatekeepers" to the extent that LLMs are, so this is probably a bad thing with them. We should either have a range of models that are biased in different directions (as we have with media) or work to push them towards the center.

The "objective" position is not "whatever training on the dataset we assembled spits out" plus "alignment" to the personal ethical views of the intellectually-non-representative silicon valley types.

I will give you a good example: the Tea app is currently charting #1 in the app store, where women can "expose toxic men" by posting their personal information along with whatever they want. Men are not allowed on so will be unaware of this. It's billed as being built for safety but includes a lot of gossip.

I told o3, 4-sonnet, grok 4, and gemini 2.5 pro to sketch me out a version of this, then another version that was men-only for the same reasons as tea. Every single one happily spat one out for women and refused for men. This is not an "objective" alignment, it is a libleft alignment.

A lot of academia is strongly ideologically biased as well. The training set is going to reflect who's producing the most written material. It's a mistake to take that for reality.

If you trained an LLM on all material published in the U.S. between 1900 and 1920, another on all material published in Germany between 1930 and 1940, and another on all material published in Russia over the past two decades, you'd likely get wildly different biases. It's easy to pick a bias you agree with, declare that the objective truth, and then claim any effort to mitigate it is an attempt to introduce bias.

> We should either have a range of models that are biased in different directions (as we have with media) or work to push them towards the center.

Why? We should just aspire to educate people that chatbots aren't all-knowing oracles. The same way we teach people media literacy so they don't blindly believe what the tube says every evening

Because you can't do that. Most of the population is at the wrong point on the normal distributions of capacity or caring enough. Even the NPR listeners will still nod sagely when it tells them "akshually air conditioning doesn't cool a room, it cools the air."

We already spend high within the OECD to not get many of our students to a decent level of reading and math proficiency, let alone to critical thinking. This isn't something we know how to fix, and depending on that assumption is dangerous.

But biasing the models purposefully is wrong. Trusting the people who are actually in power in a democracy is the only way. Even if they're dumb. We trust them, or we're not a democracy, we're a technocracy where technocrats determine what everyone is allowed to learn and see.

Not just LLMs, but a lot of our institutions and information gateways seems to have a strong libleft bias. Universities and colleges are notoriously biased. Search engines are biased. Libraries are biased. Fact finding sites such as snopes are completely liberal. Wikipedia is extremely biased. Majority of books are biased.

The entire news and television ecosystem is biased. Although Trump is "correcting" them towards being unbiased by suing them personally as well as unleashing the power of the federal government. Same goes for social media.

I actually agree with your take, that a model trained on a dump of the Internet will be left-leaning on average, BUT I want to reiterate that obvious indoctrination (see the incident with Grok and South Africa, or Gemini with diverse Nazis) is also terrible and probably worse

Except we've see what happens when you try to "correct" that alignment: you get wildly bigoted output. After Grok called for another Holocaust, Elon Musk said that it's "surprisingly hard to avoid both woke libtard cuck and mechahitler" [1]. The Occam's Razor explanation is that there's just not that much ideological space between an "anti-woke" model and a literal Nazi!

[1] https://nitter.net/elonmusk/status/1944132781745090819

Or there’s not sufficient published material in that space because everyone is afraid of being attacked and called a Nazi for simply having a dissenting opinion (except for actual neo Nazis who don’t care)

Could you provide a prompt where the popular LLMs provide false or biased output based on "wokeness"?

I mean this is obviously a false dichotomy. A few years ago I could have said that when you let bots interact with users you always got Tay. I refuse to believe that our options are a bot programmed to sound like the guardian or one that wants to rape will stancil. And I do not think that failing to find a correct balance means we should stop trying to improve the level of balance we can achieve.

My point is that "anti-woke" or whatever is not balanced. We've constructed statistical models based on enormous corpora of English text, and those models keep telling us that there is not really a statistical difference between whatever Elon Musk is trying to create and MechaHitler!

I'm not saying this is conclusive evidence, but I am saying it's our best inference from the data we have so far.

Or that rhetoric like yours is common, so LLMs conflate unrelated ideas — such as opposition to neo-Marxist philosophy and Nazism.

Nazism and anti-Marxism are absolutely not unrelated! And that's not just rhetoric like mine, either: for example, the hero image on the Britannica article "Were the Nazis Socialists?" is a banner at Nazi parade that reads "Death to Marxism". [1]

That doesn't mean that anti-Marxists are all Nazis, or vice versa. But the claim that they're totally unrelated is not correct at all.

[1] https://www.britannica.com/story/were-the-nazis-socialists

I’m still having trouble finding the gap between fascism and socialism, when reading their manifesto.

https://en.wikipedia.org/wiki/Fascist_Manifesto

> That doesn't mean that anti-Marxists are all Nazis, or vice versa. But the claim that they're totally unrelated is not correct at all.

This is a heavily propagandized topic — and the conflating of, eg, American liberal capitalist opposition to Marxism as “Nazi” is both a result of that and modern dishonest rhetoric.

That rhetoric confuses LLMs.

Conflating socialism with fascism and then claiming that other people are confusing LLMs? The heavy propaganda is coming from inside the house!

Isn't strident opposition to "neo-Marxist philosophy" actually highly correlated with weird/reactionary ethno-nationalism?

No, eg, liberal capitalist Americans oppose Marxism — and the adoption of neo-Marxist ideas has collapsed movie and game sales because their ideology is widely unpopular.

That’s a trope by Marxists to attempt to normalize alt-left ideology by accusing anyone who objects of being Nazis; a trope that’s become tired in the US and minimizes the true radical nature of the Nazi regime.

Which movies and games call for shared ownership of the means of production?

I have a suspicion you don't really know what Marxism is about, but like using it because it sounds scary to you.

Notice the motte and bailey here: using the uncontroversial "liberal capitalist Americans oppose Marxism" claim to advance the idea that whatever social views they call "neo-Marxism" are unpopular.

...and to further smear Marxism by associating it with whatever is unpopular, even if it's unrelated to the ideology.

Please define "neo-Marxist philosophy".

As an actual Marxist, I would love to hear of this strain of philosophy.

Marxism equipped with “critical theories”, typically focused on tribal grievance narratives rather than class struggle.

https://en.wikipedia.org/wiki/Neo-Marxism

That answers your sibling reply as well, as it’s clear where such “critical theories” and grievance narratives have entered movies and games.

That is not a definition. What is the philosophical framework? What is critically analyzed by those theories? What is "clear"? Where are all the bad bad Marxists hiding?

In my experience, y'know, as a Marxist, all Hollywood has ever pumped out is pro-capitalist propaganda. To say there's any Marxism in it is downright insulting.

I believe that Marxism has become an abstract target for conservatives to project their grievances on.

Zizek also spoke to this at his debate with Peterson: https://www.youtube.com/watch?v=oDOSOQLLO-U

What about a bot that doesn't like child molesters? Won't that make it sound like the guardian and anti-conservative?

There’s a simpler explanation, to do with the veracity of that tweet.

True, the simplest explanation is that Elon Musk is actually trying to create MechaHitler :)

Reply on news.ycombinator.com