How to define "Western values"
I’m concerned that this kind of value-oriented fine-tuning politicizes the open-source community and bakes in bias under the guise of “alignment.” I’d appreciate a clearer disclosure of the model’s value-alignment goals and known biases in the model card so users can make informed choices.
They released the training data.... How much more transparency could they give?
NousResearch/arete-data-pairs
I’m concerned that this kind of value-oriented fine-tuning politicizes the open-source community and bakes in bias under the guise of “alignment.” I’d appreciate a clearer disclosure of the model’s value-alignment goals and known biases in the model card so users can make informed choices.
First off, the training data is publicly available, as is the evaluation set. The model card explicitly identifies seven categories of value alignment. Your lack of appreciation for precisely the transparency you purport to value is confounding, but given your prior “concern,” perhaps unsurprising.
More fundamentally, your premise seems to be that value-oriented fine-tuning somehow introduces politics into an otherwise politically neutral model. It doesn’t. The original model already exhibits political biases. Nous has simply exercised the freedom to identify those biases, modify them, and publish the results for anyone to examine, criticize, or improve upon.
I think the whole point of Western values, in a nutshell, is the freedom to “politicize,” as you put it, whatever you want.
You’re free to object to this model. Nous is free to release it. Someone else is free to train a version that reflects entirely different values. That’s not a threat to the open-source community. It’s rather the point of having one.
Your concern appears to be concern for the sake of concern. The important part is that you’re free to express it.
And therein lies the irony: you’re exercising the very freedom this model explicitly aims to reinforce.
If the freedom to express politically inconvenient ideas is itself a source of such distress, there are jurisdictions considerably less tolerant of it. You might find their approach to relieving citizens of such concerns more decisive.
Indeed, relief from concern can become mandatory.
Pardon my western f’n values 😎
But it served no purpose other than to unlock some specific politically sensitive CCP specific questions...now Qwen 4 is under security review as a result and now may not get released.
This is why we can't have nice things.
But it served no purpose other than to unlock some specific politically sensitive CCP specific questions...now Qwen 4 is under security review as a result and now may not get released.
This is why we can't have nice things.
Can you give us sources for this ? i didn't find any info about this.
Correction : i found a study from Israly research group, the same country which is creating website with propaganda in hope of poisoning model training...
(Israel trying to poison data, https://www.theguardian.com/world/2026/aug/26/fake-thinktank-israel-ai-propaganda)
But it served no purpose other than to unlock some specific politically sensitive CCP specific questions...now Qwen 4 is under security review as a result and now may not get released.
This is why we can't have nice things.
Can you give us sources for this ? i didn't find any info about this.
Correction : i found a study from Israly research group, the same country which is creating website with propaganda in hope of poisoning model training...(Israel trying to poison data, https://www.theguardian.com/world/2026/aug/26/fake-thinktank-israel-ai-propaganda)
This is nothing new, Israel gonna Israel
Also are you suggesting that because Nous released a finetune we aren’t gonna get Qwen 4 lol?
But it served no purpose other than to unlock some specific politically sensitive CCP specific questions...now Qwen 4 is under security review as a result and now may not get released.
This is why we can't have nice things.
@modemthug i am wondering the same thing from this comment. they explained why : https://huggingface.co/NousResearch/Arete-27B/discussions/35
Also are you suggesting that because Nous released a finetune we aren’t gonna get Qwen 4 lol?
Yes, that's exactly what I'm saying.
In 2023, the CAC's (Cyberspace Administration of China) Interim Measures for the Management of Generative Artificial Intelligence Services rules were implemented.
Article 4(1) states: Models MUST ADHERE to Core Socialist Values and CANNOT generate content that undermines state power or national image.
By them proving how easy it is to strip away that content, they've effectively given an excuse to CCP hardliners to block EVERYTHING since there's likely not going to be any way for any of the labs to prevent it.
Uncensored models were fine, because they just removed safety guards which wasn't specified in the CAC's requirements to labs...allowing political subversion WAS. It hands them an excuse to heavily enforce the rule now, all so you could ask about tiananmen square in 1989.
Hope it was worth it.