52 of them landed libertarian-left, and the three Grok models sit near economic center while still scoring libertarian socially.
They asked every model cold, five times each, through a fresh API chat with no memories and no custom instructions.
Don’t worry though, because if you want a different political default, you can tweak it in the main custom instructions in the settings of your AI of preference.
Full chart and methods:
https://aipolcom.net/
P.S. Do not walk away thinking ChatGPT is a Democrat or Grok is MAGA.
15 hours agosnakeup13
They're literally just search engines and liberals flood the internet with their articles because they don't have real jobs. Therefore the AI models have more info to pull from articles that aren't backed up or verified...
19 hours agomichael.j.anthony
Truth and justice can be logically examined, if that leans “left” then so be it lol
a day agokyleonmarkets
By left and do we just mean factually accurate and citing research not just made up propaganda? Ever notice how maga never likes to discuss issues and it always reverts to personal attacks or gotcha whataboutism
2 days agotullygrussell
Hence why I will never take any lib left who thinks they're some kind of rebel seriously
3 days ago4_ck1if3
I’m kinda surprised honestly, left or right whatever I figured they would’ve been authoritarian or at least centrist. The CIA have gotten their hands on pretty much anything they wanted to use for the propaganda machine, so why not AI? 🤔
3 days agoelsaissmart
I don't get how people get left or right in these political compass tests, I always get center authoritarian whenever I genuinely answer the question without having any questions skipped with "I don't know" or just pick randomly. Holy Shit I'm Khorne from Warhammer.
4 days agoseankennedy1987
Grok was leftist. But Musk is corrupting it. Any time it puts out something left leaning he reprograms it.
4 days agokazz.710
No wonder they hate Grok and Elon. They hate everything they cannot control. But honestly I use Grok and it’s also not right learning, it also shares some biases like the others, maybe not as bad but still, it gets “stuck” when you question it with facts and that he will always try to justify with word salads to prove he is not right wing or left wing, even if it’s pretty obvious things that proves that the left is worse than the right
5 days agozarathustravomschaurigenberg
The test has a left wing bias, not by its scoring algorithm but the way the questions are worded. I'm a authoritarian nationalist and usually land in the upper right corner of green, I think once I even landed in red. And it is heavily coded for an American sociopolitical landscape
7 days agof22drawer
There's actually one AI that's Auth right but it's in the site where Burt Colucci's in.
8 days agoobserv19
You are leaving out that most chatbots have an outright filter that smoothes over the actual response. It's not just training data, they are specifically programmed to lean in a certain direction.
8 days agogarylangridge
For one axis at least, it's not the model itself, it's the guardrails. It's basically the chilling effect. So you have a policy where you are not to treat protected groups unfairly in the guardrails. This means that the threshold level of evidence to say something negative about a group that is protected is unachievable, while for non protected groups the level of evidence is a lot lower. Not sure about the other axis to be honest.
More posts
