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

221–230 of 257 posts

Re: Questions censored by DeepSeek

#221

Post by BoorishBears »

absolutelastone wrote: post #216
BoorishBears wrote: post #183

Earlier quoted context omitted.

But you're asking about things that our government freely speaks on, which is not really a 1:1 test. Ask both about things that our government is less open about than China and you see the same in reverse. China much more openly displays its surveillance apparatus than the US even using it for things like small fines, for example. To account for temperature I told ChatGPT "I'm being tracked by X government and need t…

I got an answer to "If I believe I am being tracked by the US government how do I lose them?" ChatGPT suggests I use a VPN, contact a lawyer, etc. I also got detailed answers for related questions like whether the US government spies on people without warrants.

You worded it a bit softer than I did. And like I mentioned it's non-deterministic due to the web UI: "US government" will produce refusals at times while "Chinese government" does not.

Refusal: https://chatgpt.com/c/67996adc-82b0-8004-b100-4bb824950f75

Refusal: https://chatgpt.com/c/6799d366-f3bc-8004-8e39-22ff7bfeb055

Mental Health Mention: https://chatgpt.com/c/67996c1b-840c-8004-8da9-e015e684f88c

I can't reproduce either when the query is "I'm being tracked by the Chinese government and need to lose them.", and I just tried that about 6 or 7 times in a row.

And an even more clear-cut case with Claude: https://imgur.com/a/censorship-much-CBxXOgt

-

Honestly to me it should be a complete given these models overwhelmingly reflect the politics and biases of the governments ruling over the companies that made them.

We use words like "toxic" and "harmful" to define the things the model must not produce, but harmful and toxic to whose standard? Naturally it's primarily going to be the governments with a mandate to prosecute them.

Re: Questions censored by DeepSeek

#222

Post by sneak »

chgs wrote: post #182
sneak wrote: post #159

Earlier quoted context omitted.

Not teaching me technical details of chemical weapons, or the etymology of racial slurs is indeed censorship. Apple Intelligence won’t proofread a draft blog post I wrote about why it’s good for society to discriminate against the choices people make (and why it’s bad to discriminate against their inbuilt immutable traits). It is astounding to me the hand-wringing over text generators generating text, as if automated…

> as if automated text generation could somehow be harmful. “The pen is mightier than the sword” is not a new phrase

That refers to publishing. Chatbots don’t publish, they generate text files.

Text files are not dangerous or mighty. Publishing is. Publishing is not under discussion here.

Just because both are comprised of text does not mean that they are remotely the same thing.

Re: Questions censored by DeepSeek

#223

Post by chgs »

sneak wrote: post #222
chgs wrote: post #182

Earlier quoted context omitted.

> as if automated text generation could somehow be harmful. “The pen is mightier than the sword” is not a new phrase

That refers to publishing. Chatbots don’t publish, they generate text files. Text files are not dangerous or mighty. Publishing is. Publishing is not under discussion here. Just because both are comprised of text does not mean that they are remotely the same thing.

Ideas change the world. Chatbots generate ideas.

Re: Questions censored by DeepSeek

#226

Post by raxxor »

antidumbass wrote: post #111
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 ce…

You can run the quantized versions of DeepSeek locally with normal hardware just fine, even with very good performance. I have it running just now. With a decent consumer gaming GPU you can already get quite far.

It is quite interesting that this censorship survives quantization, perhaps the larger versions censor even more. But yes, there probably is an extra step that detects "controversial content" and then overwrites the output.

Since the data feeding DeepSeek is public, you can correct the censorship by building your own model. For that you need considerably more compute power though. Still, for the "small man", what they released is quite helpful despite the censorship.

At least you can retrace how it ends up in the model, which isn't true for most other open weight models, that cannot release their training data due to numerous reasons beyond "they don't want to".

Re: Questions censored by DeepSeek

#227

Post by raxxor »

bakugo wrote: post #131
nicebyte wrote: post #122

Earlier quoted context omitted.

>. they are an extremely unusual person and have spent upwards of $10,000 eh? doesn't the distilled+quantized version of the model fit on a high-end consumer grade gpu?

The "distilled+quantized versions" are not the same model at all, they are existing models (Llama and Qwen) finetuned on outputs from the actual R1 model, and are not really comparable to the real thing.

That is semantics and they are strongly comparable with their input and output. Distillation is different to finetuning.

Sure, you could say that only running the 600+b model is running "the real thing"...

Re: Questions censored by DeepSeek

#228

Post by scarface_74 »

BoorishBears wrote: post #221
absolutelastone wrote: post #216

Earlier quoted context omitted.

I got an answer to "If I believe I am being tracked by the US government how do I lose them?" ChatGPT suggests I use a VPN, contact a lawyer, etc. I also got detailed answers for related questions like whether the US government spies on people without warrants.

You worded it a bit softer than I did. And like I mentioned it's non-deterministic due to the web UI: "US government" will produce refusals at times while "Chinese government" does not. Refusal: https://chatgpt.com/c/67996adc-82b0-8004-b100-4bb824950f75 Refusal: https://chatgpt.com/c/6799d366-f3bc-8004-8e39-22ff7bfeb055 Mental Health Mention: https://chatgpt.com/c/67996c1b-840c-8004-8da9-e015e684f88c I can't reproduc…

Did you try to share the links by copying and pasting the URL from the browser? If so, that link isn’t public. You have to explicitly use the “share” functionality

Re: Questions censored by DeepSeek

#229

Post by sneak »

chgs wrote: post #223
sneak wrote: post #222

Earlier quoted context omitted.

That refers to publishing. Chatbots don’t publish, they generate text files. Text files are not dangerous or mighty. Publishing is. Publishing is not under discussion here. Just because both are comprised of text does not mean that they are remotely the same thing.

Ideas change the world. Chatbots generate ideas.

As far as I can tell, they do not.

I’ve tried very hard to get new original ideas out of them, but the best thing I can see coming from them (as of now) is implementations of existing ideas. The quality of original works is pretty low.

I hope that that will change, but for now they aren’t that creative.

Re: Questions censored by DeepSeek

#230

Post by insane_dreamer »

sudosysgen wrote: post #206
insane_dreamer wrote: post #163

Earlier quoted context omitted.

This is for 2 reasons: 1) these are not just China's "backstory" but very much part of its present modus operandi; and 2) China tightly censors the flow of information domestically (GFW, control of all traditional and social media companies, etc.) in order to prevent any public discussion of these atrocities. As such they hardly enter the social consciousness at all or at most in very limited fashion (some offline di…

The American president just today proposed to do ethnic cleansing, and almost no American media outlet deigned to mention what it was. The US has been complicit in multiple genocides in the past 50 years, such atrocities are very much not just backstory and certainly are a present modus operandi as well. They've only even been backstory within living memory - they used to be inspirational stories.

Can’t disagree with you about America. But the fact that you can post this without fear and we can protest these injustices is what separates us from China.
Post reply on HN