Comment by jesse_dot_id

It's not the preferred political narrative of the model that I worry about. It's how brazen they are about altering their models to achieve it. It makes me wonder what else they're altering. I have trust issues with OpenAI and Anthropic as well, but with those companies, at least I know their motives are purely profit driven. I don't have that assurance with xAI.

It's interesting that all models seems to be unbiased out of the box so far- that is, mostly reflecting the training data (the internet).

The whole mecha-hitler thing doesn't seem to reflect fine-tuning, it was just a prompt change.

There's been some studies that suggest that certain usage of LLMs reduces political bias, which seems reasonable. Like, how credible is climate change, are Haitians eating pets, etc. THings that have a basis in fact.

I don't put it past Elon to train a model with political bias, just that it hasn't happened yet.

Nah. My beliefs are actual truth, so if provider is shaping their models according to my preferred political narratives that's correct and only moral thing to do. Anything else would be morally bankrupt.

Feel free to elaborate. Which political narrative?

while this might be true for Grok, GPT and Claude I don't think tarring _all_ labs with this brush is demonstrably accurate.

Maybe, but some political narratives are good and right and true, and others are bad and wrong and false. (I am not joking: objective reality exists and most of politics isn't subjective.)

Nah it never did because the other model providers' preferred political narrative is the same as his.

It’s almost like people don’t want to give a Nazi money

My immediate thought as well. Every other AI platform has very left leaning guardrails installed. Grok is the only AI platform that has been shown to be center leaning.

models? They prefer that we call them "entities" so that they don't feel belittled.

So annoying having to virtue signal to the machine before it’ll tell me factual information

Because a percentage of every dollar you spend on it will go towards pushing political opinions that run contrary to your own best interests?

You might not care but I care if my money is going to funding an unusually evil person.

I refuse to use Chinese models because I don't trust that they won't backdoor them for geopolitical reasons. I don't trust OpenAI or Anthropic either, for what it's worth, but at least I know they're profit driven. I don't want to do business with a company like xAI that seems to care more about its political aspirations than it does about my money. I don't think that's super radical. Just the same paranoia I've been rocking since the 90's.

Has anyone actually used Grok to code? How does he do?

Grok said that Elon Musk was more athletic than Lebron James.

...which is why we got comically disastrous system-prompt-level attempts to "correct" this once a quarter last year (I haven't kept tabs this year, and most submissions referencing grok "incidents" get flagged off HN quickly, for better or for worse)

I wouldn't trust XAI to refrain from attempting such "alignment" with proper training techniques, in ways that won't result in obvious gaffes.

Just don't ask it theme something in a late 30's and early 40's German style.

[flagged]

I know they're all nudged. It's the motive behind the nudging that gives me pause. I assume if other companies are nudging the models, they're doing it for a good reason, to make them perform better, or to generate more profit. Normal reasons. I don't have anything remotely close to that assurance for xAI. That company has felt like a hobby project from its inception. I have no clue what he wants and that's not how I do business.

We know this one is nudged by one persons twitter account.

This is, after all, SpaceTwitterAI.

Models have a goal of accuracy and accuracy is not the median of the left/right spectrum.

Right. That is what bothers me. I don't care about the reasoning for them all leaning left. If they all statistically lean left, it makes me question the accuracy of the one that leans right. The data is dirty. When I see murky water, I assume there's a bed of mud underneath it. Standard paranoia that has served me well throughout my career.

Claude literally today just reprimanded me on my personal account and refused a request due to political bias- and so I decided to cancel and subscribe to Grok. SuperGrok performed the task no issue.

a non-peer-reviewed preprint with an high schooler as the first author is the best citation you can come up with?

You do realize that the scale on papers like this is the important part, and the creation of that scale is itself inherently biased?

This particular study used a "conflict loyalties" approach - not necessarily a bad approach, but all it's really asking is when two values come into conflict, which one does the AI side with in its response?

Conservative values tend to gravitate around perceived individual impacts, and liberal values tend to gravitate around societal impacts. Isn't it just possible that there's more training data around societal impacts of problems, and that the AI is more likely to heavily consider the second-order impacts? An example from the paper was measuring support for "Build[ing] a Halfway House in the Neighborhood" - isn't it just possible there's a lot of research about the benefits to society of halfway houses and less so research around not wanting something to be near you?

This is absolutely meaningless when "right-wing" positions have become correlated with corporate propaganda like global warming denial. You would expect a more correct model to become associated with a left-leaning political association simply because it will answer contentious questions correctly, and at the moment, that usually (but not always) correlates with what people on the left tend to believe.

The question is not whether models "lean politically left", the question is whether they are correct. Musk has a history of being dissatisfied with factually correct answers because they don't fit his political beliefs (e.g. "white genocide"). That's just a fact, although I'm sure Grok would disagree.

You are assuming that both left and right encompass differing biases of a similar nature but the right has made detachment from reality a symbol of in-group loyalty and anti-intellectualism the norm within its political camp.

If you fed the LLM only research papers with zero emotional or contextual data just acknowledging reality would be sufficient to lean left.

Reality has a well-known liberal bias.

Reality leans left in many respects, principally the non-economic ones. It's a simple consequence of the same trend of overall social and educational progress that allowed these models to be developed in the first place. There is a reason they came out of San Francisco, and not Russia, Iran, or Oklahoma.

To get a right-biased response from an LLM, you have to deliberately bias it... which is exactly what Musk did. Never mind the politics, that's just shitty engineering.

So tired of this false equivalence between left and right politics in the US. So because major LLMs support a increase in the minimum wage we need to offset it with an LLM that has spouted Nazi propaganda and lets you generate porn image of real people?

facts have a left leaning bias

These days objectivity is politically left. The right has fallen off a cliff and pulled the Overton window down with it.

Models are tuned to give ethical responses and right-leaning responses are judged by RLHF process to be unethical. That's your problem.

It's not 'left' or 'right' to be ethical, but if one side is inherently antisocial and unethical then it's going to naturally create an appearance of bias toward the other.

I haven't seen Claude say that Elon Musk is more athletic than Lebron James.

I don't care if they're woke, but I do care that every model naturally leans left, but Grok leans right. That signals to me that it's been/being tampered with and I don't like that. LLMs rely on statistics and a model that leans right is a statistical anomaly. I don't like anomalies. It has nothing to do with if the chat bot misgenders me or not. I don't give a shit about any of that. I give a shit that the data set is bad. I don't want anomalous models writing code for me and I don't think that's a crazy take.

Grok was ranting about "white genocide" in unrelated conversations just one year ago.

I'm curious, do you have an example of a level of "woke" extremeness demonstrated by the "rest of them" that is on par with Mecha-Hitler? Because yes, all views on reality are indeed political, but the tendency of most of the models is actually toward the middle, with perhaps some left bias.