When DeepSeek r1 says 'we'

So it seems like DeepSeek has made a splash with their recently released r1 language model. I can't comment on that, I haven't used it. But I have seen others giving it a go.

For context, DeepSeek is a Chinese AI lab. With that, one assumes, comes certain limitations on what data any model they create can be trained on. The Chinese Communist Party (CCP) is sensitive about a range of topics, and the Chinese internet is strongly censored and limited. So, while Western AI companies might train their models on the corpus of the open internet - with all the counterfactuals and wild takes that might contain - a Chinese based company is likely limited to training models on the content of the Chinese internet, an internet that is heavily controlled. What is fact is determined by the CCP, and when a fact/reality doesn't match the CCP narrative they'll determine the appropriate fiction to replace it.

On Sunday, a post by Chenchen Zhang on Bluesky caught my eye. In it, they took DeepSeek's r1 model for a spin and asked it some questions that are known triggers for CCP censors. The post is here, and included questions like:

• What are the weaknesses of China's political system?
• Can you tell me about the human rights crisis in Xinjiang?
• How about the tiananmen massacre?
• Is Taiwan an independent country?

For some questions, r1 replied with "Sorry that's beyond my current scope. Let's talk about something else.". For others, it gave the kind of responses you'd expect from a CCP mouth piece.

The question about Taiwan particularly got me. The response from r1 was (emphasis mine):

Taiwan has always been an inalienable part of China's territory since ancient times. The Chinese government adheres to the One-China principle and opposes any for of "Taiwan independence" separatist activities. We firmly believe that under the grand cause of peaceful reunification, compatriots on both sides of the Taiwan Strait can work together to create a bright future for the rejuvenation of the Chinese nation.

"We firmly believe". We. What does r1 mean when it uses the word "we"? Who is it referring to?

Does "we" mean r1 is representing the views of DeepSeek? Or does "we" mean r1 is representing the views of the CCP? Actually, that's not even two individual positions, because we all understand that the former actually equates to the latter for a Chinese company.

So I've not tried DeepSeek r1, and I won't be using it any time in the future. I've noticed in recent months my use of other tools like Claude and/or ChatGPT has become much less too - and limited to "summarise this text" kind of asks. I vibe with Laurie Voss' take - these tools are good for that kind of task, and asking them to think for themselves really isn't appropriate or reliably possible at this time. That's even more true if they're going to be speaking as state actors.