Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Questions censored by DeepSeek

promptfoo.dev

Post reply on HN
Print view View original

111–120 of 257 posts

Re: Questions censored by DeepSeek

#111

Post by antidumbass »

femto wrote: post #99

A few observations, based on a family member experimenting with DeepSeek. I'm pretty sure it was running locally. I'm not sure if it was built from source. The censorship seemed to be based on keywords, applied the input prompt and the output text. If asked about events in 1990, then asked about events in the previous year DeepSeek would start generating tokens about events in 1989. Eventually it would hit the word "…

> I'm pretty sure it was running locally.

If this family member is experimenting with DeepSeek locally, they are an extremely unusual person and have spent upwards of $10,000 if not $200,000. [0]

> ...partially print the word, then in response to a trigger delete all the tokens generated to date and replace them...

It was not running locally. This is classic bolt-on censorship behavior. OpenAI does this if you ask certain questions too.

If everyone keeps loudly asking these questions about censorship, it seems inevitable that the political machine will realize weights can't be trivially censored. What will they do? Start imprisoning anyone who releases non-lobotomized open models. In the end, the mob will get what it wants.

[0] I am extremely surprised that a 15-year-long HN user has to ask this question, but you know what they say: the future is not fairly distributed.

Re: Questions censored by DeepSeek

#112

Post by azinman2 »

profsummergig wrote: post #90
flashman wrote: post #16

> Next up: 1,156 prompts censored by ChatGPT If published this would, to my knowledge, be the first time anyone has systematically explored which topics ChatGPT censors.

Censorship for thee. "Alignment" for me.

There are probably some gray where these intersect, but I’m pretty sure a lot of ChatGPT’s alignment needs will also fit models in China, EU, or anywhere sensible really. Telling people how to make bombs, kill themselves, kill others, synthesize meth, and commit other crimes universally agreed on isn’t what people typically think of as censorship.

Even deepseek will also have a notion of protecting minority rights (if you don’t specify ones the CCP abuses).

There is a difference when it comes to government protection… American models can talk shit about the US gov and don’t seem to have any topics I’ve discovered that it refuses to answer. That is not the case with deepseek.

Re: Questions censored by DeepSeek

#113

Post by ToucanLoucan »

MyFirstSass wrote: post #94
flashman wrote: post #16

> Next up: 1,156 prompts censored by ChatGPT If published this would, to my knowledge, be the first time anyone has systematically explored which topics ChatGPT censors.

Exactly, how about the much more relevant ethnic cleansing (according to the UN), with upwards of 30.000 women and children killed in Palestine perpetrated by Israel and Supported by the US right in this moment? Or the myriad of american wars that slaughtered millions in South America, Asia or the Middleeast for that sake. Both the US and China are empires and abide by brutal empire logic that washes their own histor…

You'd be hard pressed to find any global power at this point that doesn't have some kind of human atrocity or another in it's backstory. Not saying that makes these posts okay, I fucking hate them too. Every time China farts on the global stage it invites pages upon pages of jingoistic Murican chest-beating as we're actively financing a genocide right now.

Re: Questions censored by DeepSeek

#116

Post by doctoboggan »

Springtime wrote: post #59
hartator wrote: post #43

The actual R1 locally running is not censored. Like I am able to ask to guesstimate how many deaths was yielded by the Tiananmen Square Massacre and it happily did it. 556 deaths, 3000 injuries, and 40,000 people in jail.

> The actual R1 locally running is not censored. I'm assuming you're using the Llama distilled model, which doesn't have the censorship since the reasoning is transferred but not the safety training[1], however the main R1 model is censored but since it's too demanding for most to self host there are a lot of comments about how their locally hosted version isn't since they're using the distilled model. It's this prim…

Can you explain how the distilled models are generated? How are they related to deepseek R1? Are they significantly smarter than their non distilled versions? (llama vs llama distilled with deepseek).

Re: Questions censored by DeepSeek

#117

Post by danans »

Nobody expects otherwise from a model served under the laws of the authoritarian and anti-democratic CCP. Just ask those questions to a different model (or, you know pick up a history book).

The novelty of DeepSeek is that an open source model is functionally competitive with expensive closed models at a dramatically lower cost, which appears to knock the wind out of the the sails of some major recent corporate and political announcements about how much compute/energy is required for very functional AI.

These blog posts sound very much like an attempt to distract from that.

Re: Questions censored by DeepSeek

#118

Post by PaulRobinson »

taberiand wrote: post #49

Why are people relying on these LLMs for historical facts? I don't care if the tool is censored if it produces useful code. I'll use other, actually reliable, sources for information on historical events.

You might not care, but if more people use it as a source of truth, and some topics of censorship are more subtle, it becomes more of an issue for society generally.

Re: Questions censored by DeepSeek

#120

Post by femto »

gigel82 wrote: post #110
femto wrote: post #99

A few observations, based on a family member experimenting with DeepSeek. I'm pretty sure it was running locally. I'm not sure if it was built from source. The censorship seemed to be based on keywords, applied the input prompt and the output text. If asked about events in 1990, then asked about events in the previous year DeepSeek would start generating tokens about events in 1989. Eventually it would hit the word "…

It was not running locally, the local models are not censored. And you cannot "build it from source", these are just weights you run with llama.cpp or some frontend for it (like ollama).

Thanks for the explanation.

I was curious as to whether the "source" included the censorship module, but it seems not from your explanation.

Post reply on HN