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

191–200 of 257 posts

Re: Questions censored by DeepSeek

#191

Post by BeetleB »

lovich wrote: post #143
BeetleB wrote: post #139

I have to mirror other comments: I find the obsession with Chinese censorship in LLMs disappointing. Yes, perhaps it won't tell you about Tiananmen square or similar issues. That's pretty OK as long as you're aware of it. OTOH, a LLM that is always trying to put a positive spin on things, or promotes ideologies, is far, far worse. I've used GPT and the like knowing the minefield it represents. DeepSeek is no worse, a…

> That's pretty OK as long as you're aware of it. We’re only aware of it because people obsess over it. If you didn’t have censorship hawks or anti China people beating the drum about Tiananmen Square, how likely would it be that anyone outside of China actually discovered the model wouldn’t talk about that. Even your example about putting a positive spin on things or promoting ideologies. When I read ChatGPT 3s outp…

> We’re only aware of it because people obsess over it.

Maybe true for you, but not for me. My operating assumption is all LLMs are "censored".

> If you didn’t have censorship hawks or anti China people beating the drum about Tiananmen Square, how likely would it be that anyone outside of China actually discovered the model wouldn’t talk about that.

The more relevant question is "If an average user didn't know that Chinese LLMs don't talk about it, how detrimental is that?"

And my response is "Probably not more than the existing LLMs in use." Easy to bet money that 99.99% of all DeepSeek usage is not impacted by that censorship.

I'm not saying one shouldn't highlight it at all. It doesn't merit being mentioned in every DeepSeek thread I see on the Internet. Just as we don't talk about AI Safety in threads related to using GPT4 for solving math problems.

Put another way, if my company is using Chinese factories, does it make sense for me to highlight "Yes, but can those Chinese factory workers discuss Tiananmen Square?" in every manufacturing related meeting?

Re: Questions censored by DeepSeek

#192

Post by hi_hi »

Although the ability to censor is somewhat interesting and important to understand at a technical level, the amount of pearl clutching and fear mongering going around in traditional media about DeepSeek is extraordinary.

Even so called independent publications are showing extreme bias. Not once does the concept or word "hallucination" appear here, now it's "misinformation". And all these concerns about submitting personal information, while good advice, seem strangely targeted at DeepSeek, rather than any online service.

https://www.theguardian.com/technology/2025/jan/28/experts-u...

Sigh, I'm not even mad, just disappointed at how politicised and polarising these things have become. Gotta sell those clicks somehow.

Re: Questions censored by DeepSeek

#193

Post by SequoiaHope »

hi_hi wrote: post #192

Although the ability to censor is somewhat interesting and important to understand at a technical level, the amount of pearl clutching and fear mongering going around in traditional media about DeepSeek is extraordinary. Even so called independent publications are showing extreme bias. Not once does the concept or word "hallucination" appear here, now it's "misinformation". And all these concerns about submitting per…

Perhaps a minor point but hallucination was never a good description for errors produced by the model - all responses, correct or incorrect, are in essence hallucinations.

Re: Questions censored by DeepSeek

#194

Post by ngcazz »

patrickmay wrote: post #158

It's interesting to see the number of comments that consist of whataboutism ("But, but, but ChatGPT!") and minimization of the problem ("It's not really censorship." or "You can get that information elsewhere."). I like to avoid conspiracy theories, but it wouldn't surprise me if the CCP were trying to make DeepSeek more socially acceptable.

I suspect this is mainly a regulatory issue

Re: Questions censored by DeepSeek

#195

Post by Springtime »

femto wrote: post #190
Springtime wrote: post #174

Earlier quoted context omitted.

Only the R1 671B model (aka just plain 'R1') has the censorship being discussed in the article. The smaller parameter models are fine-tunings of Llama and Qwen, and the former at least doesn't have the censorship. This has caused a lot of conflicting anecdotes since those finding their prompts aren't censored are running the distilled/fine-tuned models not the foundational base model. A sibling comment was facetiousl…

Is it true to say that there are two levels of censorship at play here? First is a "blunt" wrapper that replaces the output with the "I am an AI assistant designed to provide helpful and harmless responses" message. Second is a more subtle level built into the training, whereby the output text skirts around certain topics. It is this second level that is being covered by the "1,156 Questions Censored by DeepSeek" art…

The Deepseek hosted chat site has additional 'post-hoc' censorship applied from what people have observed, if that's what you're referring to. While the foundational model (including self hosted) has some just part of its training which is the kind the article is discussing, yes.

Re: Questions censored by DeepSeek

#196

Post by blackeyeblitzar »

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 "…

There are several people who have claimed that they ran into political censorship, even offline. It seems like the experience is different based on which specific model you’re using.

Re: Questions censored by DeepSeek

#197

Post by blackeyeblitzar »

A_D_E_P_T wrote: post #30
bbqfog wrote: post #18

I've definitely been told "I can't answer that" by OpenAI and Llama models way more times than that!

Claude is the worst. It can barely even tell jokes. In terms of "openness" and "willingness to respond" I rank 'em: Deepseek > Chat-GPT = Llama >>> Claude. Deepseek seems like a neutral tool. Claude is very, very preachy. I'm happy that there doesn't appear to be any reason to ever use it again.

I agree that Claude can be very preachy and frustrating. But I disagree that DeepSeek is the most neutral. I think it is actually the least neutral because the censorship is by design and forced by the government. Claude basically has clumsy safety features that are not sophisticated to stay out of the way but without malicious intent, unlike DeepSeek.

Re: Questions censored by DeepSeek

#198

Post by blackeyeblitzar »

BeetleB wrote: post #191
lovich wrote: post #143

Earlier quoted context omitted.

> That's pretty OK as long as you're aware of it. We’re only aware of it because people obsess over it. If you didn’t have censorship hawks or anti China people beating the drum about Tiananmen Square, how likely would it be that anyone outside of China actually discovered the model wouldn’t talk about that. Even your example about putting a positive spin on things or promoting ideologies. When I read ChatGPT 3s outp…

> We’re only aware of it because people obsess over it. Maybe true for you, but not for me. My operating assumption is all LLMs are "censored". > If you didn’t have censorship hawks or anti China people beating the drum about Tiananmen Square, how likely would it be that anyone outside of China actually discovered the model wouldn’t talk about that. The more relevant question is "If an average user didn't know that C…

I think the reason it is relevant is because there are a lot of people flocking to the DeepSeek app who don’t know about these things. And discussing this makes it clear that there can be subtle manipulation and propaganda if they rely heavily on this app. TikTok has the same vulnerability since it is also subject to the same power of the CCP, but we saw that millions use it without awareness. There’s a chance to build that awareness now when the censorship is obvious instead of difficult to detect.

Re: Questions censored by DeepSeek

#199

Post by zb3 »

patrickmay wrote: post #158

It's interesting to see the number of comments that consist of whataboutism ("But, but, but ChatGPT!") and minimization of the problem ("It's not really censorship." or "You can get that information elsewhere."). I like to avoid conspiracy theories, but it wouldn't surprise me if the CCP were trying to make DeepSeek more socially acceptable.

I'm not from CCP, but since I don't live in China, Chinese "alignment" is not really a concern for me.

Is chinese-specific censorship a concern for you?

Re: Questions censored by DeepSeek

#200

Post by zb3 »

danans wrote: post #117

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 p…

Obviously. What else could Chinese engineers do? The most they can do is to convince the company/party to make the model open source and it seems they've done that.
Post reply on HN