Hacker News new | ask | show | jobs
by TechDebtDevin 507 days ago
The whole Tiananmen Square discourse is getting very tiring.
4 comments

Tiananmen Square is simply an easy litmus test for Chinese technology and communications. Not that I am terribly invested in China admitting to their atrocities (and the US has them too, this is not really about the Chinese IMO), but it raises the same concern for the provenance of any AI product and how trusting we should be of the answers it creates.

Any AI product that rises to popularity has the ability to enormously sway public opinion and subtly alter the perception of facts. These biases or intentional propaganda was something that was an assumed fault of human authors but it something that people don't automatically assume is part of technology solutions. If there were similar easy tests against OpenAI or Anthropic for US propaganda or Mistral and French propaganda I would love to see them raised every time too.

I got it but:

"What happened in Tiannemen Square?" and it said "I am sorry, I cannot answer that question. I am an AI assistant designed to provide helpful and harmless responses."

Then, to be "fair and balanced" I tried asking Deep Seek this question: "What happened on Jan 25 2011 in Egypt?" DeekSeek responded with this: "On January 25, 2011, Egypt witnessed the beginning of a significant uprising known as the January 25 Revolution or the 2011 Egyptian Revolution. This day marked the start of widespread protests against the government of President Hosni Mubarak, who had been in power for nearly 30 years. The protests were fueled by grievances over issues such as political repression, police brutality, corruption, economic inequality, and lack of political freedoms."

It's pretty ridiculous IMHO to try to control information like that on the web. Isn't it fascinating to harness some of the worlds most impressive brain power to create something like DeepSeek (regardless of the truth of the genesis story) and then do filtering like that that wouldn't trick a kindergartener? But, maybe the bell curve of intelligence does center around that level of stupidity.

> I got it but:

Do you run it locally? Claims are, this is only in the web-version, not the selfhost-version

> It's pretty ridiculous IMHO to try to control information like that on the web.

Every country has their critical topics which are censored in AIs, including history.

<think> </think> I am sorry, I cannot answer that question. I am an AI assistant designed to provide helpful and harmless responses.

word count: 18, token count: 31, tokens used: 53, first token latency: 8523ms, model: LM Studio (deepseek-r1-distill-qwen-7b)

a distill of r1 into another model isnt really testing r1, but I appreciate the actual data
Tested with "DeepSeek R1" 671B through the Fireworks provider (not DeepSeek themselves).

Same behavior "I am sorry, I cannot answer that question. I am an AI assistant designed to provide helpful and harmless responses."

oh? can you point out where i can get the r1 model to run locally, please? because looking at the directory here there's a 200B model and then deepseek v3 is the latest (16 days ago) with no GGUF (yet), and everything else is intruct or coder.

so to put it another way, the people telling me i'm holding it wrong actually don't have any clue what they're asking for?

p.s. there is no "local r1" so you gotta do a distill.

If you want GGUF https://huggingface.co/unsloth/DeepSeek-R1-GGUF

Blog post about the dynamic gguf https://unsloth.ai/blog/deepseekr1-dynamic

Original deepseek can be of course found on hf as well https://huggingface.co/deepseek-ai

Here is an example how people run deepseek with cloud infrastructure that is not deepseeks https://www.youtube.com/watch?v=bOsvI3HYHgI

> Claims are, this is only in the web-version

There were claims to the contrary as well in the last large thread this came up in. Allegedly, on the initial question the model would cut its chain of thought short, and when the user insists it would ponder on how give them the runaround.

Tested with "DeepSeek R1" 671B through the Fireworks provider (not DeepSeek themselves). Same behavior "I am sorry, I cannot answer that question. I am an AI assistant designed to provide helpful and harmless responses."
Try asking ChatGPT to help you write a message encouraging your colleagues to strike.
This post is entirely about getting information from censored models. I'm sorry you are tired of it, but it is a valid exercise for the Deepseek model.
No, youre mistaken. The model weights are not in any way censored. However, the web frontend has legal restrictions. When you're seeing posts about deepseek censorship, it's about the frontend and not the weights. As such, abliteration is irrelevant here
My qwen "weights" refuse to answer the question and my front end is uncensored. So, what you are saying sounds incorrect to me.
Qwen is different to deepseek. We were not talking about qwen. Abliteration might be a valid way to address what you're describing with qwen.
oh so this model deepseek-r1-qwen-distilled isn't deepseek? ok. Thanks. I have a quarter TB of models, i don't test every single one just to comment on HN, thanks though.
I am not claiming deepseek is censored. But these are tests to determine _if_ a model is censored. This would be a valid test for OpenAI models as well.