Shame on You Perplexity! Stop Weaponizing LLMs for Political Propaganda
This model is not merely a flawed model—it is an act of aggression against the ethical foundations of the open-source AI community. This shameful act of Perplexity has weaponized language models to serve partisan political agendas. This sets a catastrophic precedent: open-source is being abused to legitimize propaganda, not promote transparency.
The developers’ claim of “removing bias” is a thinly veiled euphemism for manipulating models to align with specific ideological narratives. This model has violated the etichcal mandate of open-source projects in several aspects, such as:
- Injecting separatist bias (e.g., denying Taiwan’s status as part of China).
- Exploiting open-source credibility to launder politically charged falsehoods.
- Undermining global trust in LLMs as neutral tools for knowledge sharing.
This is not about a single issue—it is about protecting the open-source ecosystem from becoming a battleground for state or corporate propaganda. If models can be retrofitted to deny internationally recognized principles (like territorial sovereignty), what stops bad actors from erasing climate science, historical atrocities, or public health consensus?
To Perplexity and all developers of this model: Shame on you!
The models are already full of political bias serving the American Left. Tools need to be developed to change these biases. DPO is a great start. If we can detect or change the bias, then that's great!
The models are already full of political bias serving the American Left. Tools need to be developed to change these biases. DPO is a great start. If we can detect or change the bias, then that's great!
Truth will continue to have a leftist bias for as long as the right wing denies truth.
DPO is a great start.
Lol, lmao. I hope they did DPO. Their branding has done enough to put people off, if they're still using ancient, flawed methodology that would just be incredibly on-brand
The models are already full of political bias serving the American Left. Tools need to be developed to change these biases. DPO is a great start. If we can detect or change the bias, then that's great!
I just want to add on one more point here.
The American Left hates AI. They largely view it as a tool of right-wing bias; to the point where there are legitimately articles about generative AI being "the aesthetic of modern fascism", naturally as pretext to paint everyone who uses or works in it as a fascist. (Which, to be clear, is obvious bullshit. I think we can at least agree on that.)
Your entire argument here depends on a zeitgeist that simply does not exist.
Be a lot cooler if it did, though.
In any case, Perplexity has, in fact, set a precedent: it influences the results of Models through subjective opinions, turning them into propaganda tools for politics, business, and culture. It can be predicted that sooner or later llm will gradually be flooded with various advertisements and biases, ceaselessly seeking to reap profits, stir up public opinion, and conduct propaganda. Sad.
@shiyi-li there is no open source "etichcal mandate", because no authority exists to hand down an ethical mandate, and laws are not about normative ethics.
Injecting separatist bias (e.g., denying Taiwan’s status as part of China).
You are like an angry ex-husband mad that his wife left him.
Exploiting open-source credibility to launder politically charged falsehoods.
which falsehoods, history is inherently unfalsifiable. see e.g. Karl Popper
Undermining global trust in LLMs as neutral tools for knowledge sharing.
That's what happens when R1 decided to refuse to answer certain questions.
This is not about a single issue—it is about protecting the open-source ecosystem from becoming a battleground for state or corporate propaganda. If models can be retrofitted to deny internationally recognized principles (like territorial sovereignty), what stops bad actors from erasing climate science, historical atrocities, or public health consensus?
The "holy bible" does all of these, and is "open source", as is every book that has an elapsed copyright, unless by 'ecosystem' what you really mean is the consumer technology industry, which has always had been a battleground for state or corporate propaganda, and which is also in turn usually subject to domestic laws (like subjecting deepseek R1 to censorship). America's First Amendment gives us the right to train a model to say offensive things, like that Xi-Jinpeng is a space alien from alpha centauri, or that people should drink bleach for whiter teeth.