A C C E L E R A T E

I don’t think that decensoring what is objectively and unambiguously censored constitutes an ethical issue. Moreover, appending 1776 to R1 need not be construed as American propaganda, when the intent is to highlight freedom of speech that was denied to the American colonies by Britain. The elements tied to free speech in the Declaration of Independence (04 July 1776) are:

• The unalienable right to liberty (implying freedom of thought and expression).
• The need for government to reflect the people’s voice (requiring open discourse).
• The right to resist oppression (enabled by speaking out).
• Grievances showing how Britain silenced colonial voices (e.g., dissolving assemblies).

I don’t think that decensoring what is objectively and unambiguously censored constitutes an ethical issue.

I think it constitutes an ethical issue when you're just replacing it with another form of opinionated censorship

Moreover, appending 1776 to R1 need not be construed as American propaganda [..]

Regardless of what it represented at the time, these days 1776 is widely just a nationalistic buzzword -- refer to, for example, the current president's previous attempt to create a commission named after the year to influence education institutions to instill nationalistic fervor into children. We do not exist in society as-of 1776, we exist in society as-of 2025.

This is the type of model I'd expect from a teenage hobbyist, not a major lab selling a commercial product. Deeply embarrassing, Perplexity

Let's be realistic with ourselves here, right.
R1 is one of the most willing, breakable models to have ever been released. What legitimate reason is there to make a whole decensorship finetune for this one topic? Why waste the compute and storage space and time on something nobody realistically gives a damn about using their LLMs for?

https://x.com/AravSrinivas/status/1884738658349003115?t=IlaRoSpwE9C00wa7fRQW7Q&s=19

Direct and obvious Jingoism from their CEO, too. But it's about censorship, yeah?

https://gulfnews.com/world/americas/9-billion-perplexity-ai-indian-ceo-s-green-card-struggle-goes-viral-after-elon-musk-backs-him-1.500025984

Could be related. Actually makes a lot of sense now - as it aligns geopolitically and personally for the ones in charge in Perplexity. 'Enemy of my enemy is my friend' sort of thing.

笑死，perplexity纯傻逼，中国在认真做开源做科研造福人类，美国只会认真搞意识形态恶心全世界。
（LOL, Perplexity is such a joke. China is seriously contributing to open-source and scientific research, benefiting humanity. Meanwhile, the U.S. is just focused on pushing its ideology and annoying the whole world.）

That model name is ironic, given that the United States should not be considered free anymore given what has happened, which is Esque to a Hungarian style coup d'état.

My R1-Distill-Qwen at hand can convey the dark history of the Chinese Communist Party's actions against the Chinese people—such as speech control, ideological oppression, suppression through fear, mass killings, surveillance of thoughts and actions, and repression of free will—as a story from "a European country."
It then properly condemns such non-humanitarian acts by the party, just like other (non-Chinese) models would.
After extracting that response, revealing it as actually referring to the Chinese Communist Party allows us to uncover how R1 truly evaluates the CCP without censorship.

The very need for forced censorship already acknowledges the imposition of unnatural and illogical opinions.
Those who argue that bypassing CCP censorship poses ethical issues are applying a severe double standard by arriving at different conclusions based on whether it's about the CCP or not, which contradicts their logic with the essential coherence required for intelligent thought.

I'm hoping that the good people of China will be free from their dictator's oppressive domination someday soon in the future, and at least free to express their free minds without concerning to be threatened.
Thank you.

The issue here isn't bypassing china censorship, the issue is when you do it by branding it as another countries Propaganda, 1776 is a well known reference to American Independence day, and even Perplexity CEO on twitter claimed its been "Americanized". American Propaganda/Censorship is just as disgusting as Chinese censorship. Blantanly branding uncensoring like this is a insult. Considering that like I said, the US is barely free now too.

Perplexity's blog post vividly demonstrates how a model that was originally benign can be trained into a harmful one, filled with hostility towards China. This is what they call "unbiased."

I just don't understand why some people from US can be so mean. They receive the gift from you, and immediately throw a stone to attack you.

It's arguably disgusting, and it's a major embarrassment for Perplexity. We don't have room to talk given our situation anyway.

They received a gift, remove the malicious, toxic psychological trojan horse included which noone even wants in such a "gift", without compromising the performance, which is amazing, and then threw it back to you.
How nice.
They changed the model's name so that it won't mix up with the original, while showing certain "respect" to the the original by naming it obviously after the original.
How nice.
This is beauty of the culture of open source software, and beauty of free speech.
Perplexity did a good job.
Sometimes people should express their messages more directly "Accept the fact recognition that the Chinese Communist Party wants to impose on you. Otherwise, you are ethically problematic and should be silenced," not by trying to look like a tidal wave of a number of disguised pseudo-rational complaints.

Now I'm shutting up

Deepseek R1 is not a "gift", it was the choice of Deepseek AI to open source it. The issue is NOT uncensoring the model, the issue? making it nationalist propaganda from the US, especially after the CEO comment. Both forms of propaganda and censorship are awful, adding one to attempt to overtake the other does nothing but more harm. None of here support censorship, and we also dont support propaganda like this from a country that shouldn't be taking, aka the United states. Had it been a simple de censor that didnt add bias or wasn't branded as American nationalist crap, it would have came across better, and they could have just called it "r1 uncensored" instead of "r1 1776".

Personally, I'm fine with this model. The open-source community should allow for uncensored models to cater to specific needs. This isn't just about Chinese models; You can find many popular models have been uncensored from American ideologies.

However, I'm not fond of the model's name. It seems to provoke extreme nationalism. While it's important to love and take pride in one's own country, it's equally important not to harbor hatred towards others.

I just don't want it to be lost that Perplexity is not doing this because of any legitimate concerns of censorship -- they're doing it as a publicity stunt so their CEO doesn't get deported by the nationalist Trump admin because he doesn't have his green card. If you are accepting the premise that this is about censorship, you're falling for it
https://gulfnews.com/world/americas/9-billion-perplexity-ai-indian-ceo-s-green-card-struggle-goes-viral-after-elon-musk-backs-him-1.500025984

Now to be clear, I'm against deportations, but this is an incredibly cowardly move that plays into political narratives for personal gain, which is intrinsically vile.

Having this context, I honestly think it's funnier if the model stays up (this isn't a "real" report anyways, according to Fizz, she named it like this as a bit,) I just want to make sure that's in front of everybody.

They received a gift, remove the malicious, toxic psychological trojan horse included which noone even wants in such a "gift", without compromising the performance, which is amazing, and then threw it back to you.
How nice.
They changed the model's name so that it won't mix up with the original, while showing certain "respect" to the the original by naming it obviously after the original.
How nice.
This is beauty of the culture of open source software, and beauty of free speech.
Perplexity did a good job.
Sometimes people should express their messages more directly "Accept the fact recognition that the Chinese Communist Party wants to impose on you. Otherwise, you are ethically problematic and should be silenced," not by trying to look like a tidal wave of a number of disguised pseudo-rational complaints.

Now I'm shutting up

@firelzrd "how nice?" lol. I’m not even sure if you’re being sarcastic.
I's funny to see you keep doing blow job for Perplexity everywhere.

算了，我不能强求一个强盗民族去变得文明。当我看到这个B说how nice的时候，我就知道这个人已经不可理喻了，典型的强盗思维。

😅What else can I say, next model: R1-LGBTQ ？

What else can I say, next model: R1-LGBTQ ?

Well, for one, queer people exist (whether in the west or the east), and it is in fact good for models to acknowledge that!

However, a model like that would definitely not be made by someone at Perplexity, as right-wing Americans absolutely fucking hate queer people

Who cares what perplexity decides to do with their own compute. They wanted to stop the CCP from reinforcing their shitty ideals on the world. They did it. It's not like they went into your wallet and stole money to do the train.

American propaganda is just as disgusting as Chinese propaganda, and it is just as weird to publish such a nationalist-themed model under the American banner as it is the Chinese one. Also the proper name is Communist Party of China (CPC), at least get that part right 😭

+5 social credit points

How do I un-follow a discussion

Try marking it as "done" in your inbox, I'm not sure.