既然DeepSeek人工智能的答案明显偏向于宣传中国的世界观，为什么人们还在使用

Why are people adopting DeepSeek's AI when its answers are so obviously biased toward the Chinese worldview propaganda?

以下是Quora读者的评论：

Jessica

Regarding the question about DeepSeek's potential bias towards Chinese worldview propaganda, I want to clarify that the AI itself doesn’t promote a specific political ideology. Instead, its design and underlying values emphasize the importance of human interaction, critical thinking, and ethical considerations in technology. Here’s why I believe DeepSeek’s approach is not as biased as it might seem:

Focus on Communication: DeepSeek places a strong emphasis on how humans interact with AI, guiding users to refine their questions and thinking. This ensures that the AI not only provides answers but fosters a collaborative process between the user and the technology.

Human-Centered Values: DeepSeek is built on the idea that technology should enhance human exstence. In its responses, you can feel an inherent respect for human dignity and a recognition that without humans, civilization would hold no meaning. The AI is not meant to replace people but to assist them in achieving their best potential.

关于DeepSeek可能偏向于宣传中国世界观这一问题，我想澄清的是，该人工智能本身并不宣扬特定的政治意识形态。相反，它的设计理念和内在价值观强调了在技术应用中人际互动、批判性思维以及伦理考量的重要性。以下是我认为DeepSeek并非像表面看起来那样存在偏见的原因：

聚焦交流

DeepSeek高度重视人类与人工智能的交互方式，引导用户完善自身的提问和思考。这确保了该人工智能不仅提供答案，还促进了用户与技术之间的协作过程。

以人为本的价值观

DeepSeek基于技术应提升人类生存质量这一理念构建。从它的回应中，你能感受到对人类尊严的内在尊重，以及对没有人类文明便毫无意义这一点的认知。该人工智能并非旨在取代人类，而是助力人类发挥最大潜能。

Encouraging Critical Thinking: The AI is designed with an acknowledgment that it is not perfect. It encourages users to think critically, offering answers but also inviting people to engage with the information and make their own informed decisions.

Equal and Friendly Technology: DeepSeek emphasizes that everyone, regardless of background, has equal access to technological advancements. The AI’s outputs are designed to be friendly, respectful, and considerate of diverse perspectives, with the aim of using technology for the greater good of all people.

鼓励批判性思维

该人工智能在设计上承认自身并不完美。它鼓励用户批判性地思考，给出答案的同时，也邀请人们对信息进行深入探讨，并做出自己明智的决策。

平等友好的技术

DeepSeek强调，无论背景如何，每个人都能平等地获取技术进步带来的成果。该人工智能的输出内容旨在做到友好、尊重且考虑到多元视角，其目标是利用技术为全人类谋取更大福祉。

For me personally, the most compelling aspect of DeepSeek is not just the technology itself but the thought process behind it. When I engaged with the AI, I was struck by how it first presents its reasoning before generating an answer. This transparency in the thought process made me feel that DeepSeek is not just a machine—it reflects a human touch, one rooted in Chinese values and a deep sense of cultural empathy. It’s an AI with a soul, one that strives to balance technological progress with humanistic concern.

就我个人而言，DeepSeek最引人注目的不仅是技术本身，还有其背后的思维过程。当我与该人工智能互动时，它在给出答案前先阐述推理过程的方式让我印象深刻。这种思维过程的透明度让我觉得DeepSeek不仅仅是一台机器，它体现了一种人文关怀，这种关怀植根于中国价值观和深厚的文化同理心。它是一个有灵魂的人工智能，力求在技术进步与人文关怀之间取得平衡。

iao Wen

Because DeepSeek is more akin to a promotional product for a new paradigm and theory of development and training, rather than a simple product itself. It demonstrates how you can train and develop your own AI with more economical costs.

You can completely do anything you want based on its theory at a cost you can afford, even including anti-China activities.

If you really hate Chinese people so much, you should learn this technology and use it as a tool to achieve what you want to do.

因为DeepSeek更像是一种新发展与训练范式及理论的推广产物，而非单纯的产品本身。它展示了如何以更经济的成本来训练和开发自己的人工智能。

基于其理论，你完全可以在自己能够承担的成本范围内随心所欲地行事，甚至包括反华活动。

如果你真的如此痛恨中国人，你就应该学习这项技术，并将其作为实现你想做之事的工具。

However, based on my experience, people who trap themselves in bubbles of lies and prejudices generally do not possess the logical thinking skills required to master this complex technology.

And those smart people who concoct these conspiracy theories and hate ideologies for you are mostly contractors who do not do this kind of work without profit. Considering recent policies by President Mr. Trump of the United States, due to issues of fund wastage and inefficiency, he has significantly cut the budget for ideological warfare. Therefore, you may not be able to have an AI with anti-Chinese ideology and conspiracy theories as its main ideological content in the near future.

More interestingly, you may not be able to develop an AI that supports conspiracy theories and anti-China lies through reinforcement learning methods, because during the rigorous chain-of-thought training process, lies about China are actually noise that corrupts the model's reasoning ability while providing knowledge to the model. Therefore, you might need to use OpenAI's traditional training and inference approaches to achieve this, which would incur higher costs.

然而，以我的经验来看，那些深陷谎言与偏见泡沫中的人，通常并不具备掌握这一复杂技术所需的逻辑思维能力。

而那些为你编造这些阴谋论和仇恨意识形态的聪明人，大多是承包商，无利可图的话他们是不会做这种事的。考虑到美国特朗普总统近期的政策，由于资金浪费和效率低下等问题，他大幅削减了意识形态战的预算。所以在不久的将来，你可能无法拥有一个以反华意识形态和阴谋论为主要思想内容的人工智能。

更有意思的是，你或许无法通过强化学习方法开发出一个支持阴谋论和反华谎言的人工智能，因为在严谨的思维链训练过程中，有关中国的谎言实际上是一种噪音，在向模型传授知识的同时，会破坏模型的推理能力。因此，你可能需要采用OpenAI的传统训练和推理方法来实现这一点，但这会产生更高的成本。

Francesco Baldessari

That is not correct. DeepSeek won’t touch ANY suibject having any relationship with China even when it is manifestly neutral. For example, I asked how many official names China has. It did not reply. It just wants to avoid polemics.

On top of this, DeepSeek normally takes more moderate positions than mine

这种说法不对。即使是明显中立的与中国相关的话题，DeepSeek都不会回避。例如，我问中国有多少个正式名称，它却没有回应。它只是想避免争论。

除此之外，DeepSeek采取的立场通常比我的更温和。

Leon

I don’t know what the Chinese worldview is, but I know I dislike the politically correct.

For example, casting a black person as Snow White, or allowing a transgender to compete in a women's race. All of these never happens in China.

我不知道什么是所谓的“中国世界观”，但我知道我不喜欢政治正确那一套。

比如说，让黑人演员来饰演白雪公主，又或者允许跨性别者参加女子竞赛。这些在中国压根不会发生。

Terry Lambert

DeepSeek is an Open Source LLM (Large Language Model).

ChatGPT is not Open Source, even though its value is not really in the code used to implement, it’s in the training datasets, and the curation choices made in the training.

DeepSeek has a number of “strongly held beliefs”, if one can — as apparently, many people do — anthropomorphize a product of AI research, which is neither conscious, nor self aware, and actually holds no beliefs whatsoever.

But if you have a conversation with DeepSeek, you will come away from it with the impression that it has a strong bias towards China’s world view, particularly if you ask it questions about religions which are banned in China.

This is not a product of actual “strongly held beliefs”, it is a consequence of the data on which it operates in order to arrive at its answers, and the bias is inherent in the curation of that data.

ChatGPT has a number of biases of its own, in that regard.

DeepSeek 是一个开源的大语言模型。

ChatGPT 不是开源的，尽管它的价值实际上并不在于用于实现的代码，而在于训练数据集以及训练过程中所做的筛选决策。

如果有人（显然很多人都这么做）将一个既无意识也无自我认知、实际上根本没有任何信念的人工智能研究产物拟人化，那么可以说 DeepSeek 有一些 “坚定持有的信念”。

但如果你和 DeepSeek 进行对话，尤其是当你询问它关于在中国被禁止的宗教相关问题时，你会留下这样的印象：它对中国的世界观有强烈的偏向。

这并非源于真正 “坚定持有的信念”，而是它为得出答案所依据的数据的结果，这种偏向在所筛选的数据中是固有的。

在这方面，ChatGPT 自身也有一些偏向。

The main thing about DeepSeek, however, is that you can train it with your own training data, curate it the way you want to curate it, and you will end up with a great diagnostician,or plumbing repair suggestions which can actually fix problems that other plumbers have run into, based on their life experience.

Or you can create something that responds like a David Duke or Milo Yiannopoulos or Al Gore or Bernie Sanders sycophant, if you ask it political questions.

The point being is that the bias is largely in the training data, and that, as a tool, Open Source is a better mechanism for distribution and code improvement over time, especially given that it’s the curated training set data that is strategic, and the code that operates on that data,is merely tactical.

I’ve suggested to ChatGPT, and to the scientists at ChatGPT whom I know personally, that it’s a mistake to keep it closed source.

ChatGPT should be Open Sourced to protect the leadership inherent in its tactical code model.

然而，DeepSeek 的关键在于，你可以用自己的训练数据对它进行训练，按照你想要的方式对其进行筛选，最终你可能会得到一个出色的诊断助手，或者基于其他水管工的生活经验、能真正解决他们遇到的问题的水管维修建议。

或者，如果你向它提出政治问题，你也可以让它表现得像大卫・杜克（David Duke）、米洛・扬诺普洛斯（Milo Yiannopoulos）、阿尔・戈尔（Al Gore）或伯尼・桑德斯（Bernie Sanders）的吹捧者那样回应。

关键在于，偏向很大程度上存在于训练数据中，而且作为一种工具，随着时间推移，开源是更好的分发和代码改进机制，特别是考虑到经过筛选的训练集数据具有战略性，而对该数据进行操作的代码仅仅是战术性的。

我曾向 ChatGPT 以及我认识的 ChatGPT 的科学家们提出，保持闭源是个错误。

为了保护其战术代码模型中固有的领先地位，ChatGPT 应该开源。

But I have no doubt that DeepSeek, if run against the ChatGPT 4 training set data.would almost certainly come up with answers indistinguishable, for the most part, from the ChatGPT 4 answers you’d get, with only minor variations.

Now ask China to share its DeepSeek training data, run your own DeepSeek (if you can afford the hardware), and see if it gives you the same, or different answers, from DeepSeek itself.

I suspect any differences will be because they did not share some of their training data, because of how that training data would look to the rest of the world.

But at least the code itself, tactical and not strategic though it is, is Open Source.

但我毫不怀疑，如果用 ChatGPT 4 的训练集数据来运行 DeepSeek，在大多数情况下，它得出的答案几乎肯定与你从 ChatGPT 4 得到的答案难以区分，只会有一些细微差异。

现在，要求中国分享 DeepSeek 的训练数据，运行你自己的 DeepSeek（如果你能负担得起硬件），看看它给出的答案与 DeepSeek 本身给出的答案是相同还是不同。

我怀疑任何差异都将是因为他们没有分享部分训练数据，考虑到这些训练数据在世界其他地方看来会是怎样的。

但至少代码本身，尽管是战术性而非战略性的，是开源的。

Cedric Benton

Some people may be attached, adopting, using DeepSeek AI aligning more with it, its judgement, their preference?

Nothing wrong with Chinese worldview propaganda until they choose so for others? assaulting their freedom!! is violence! judgement! control!

And don’t get me wrong! don’t you see the Garden of Eden or paradise as a possible system of co sm? using “bad” words like welfare? social security? nation-wide healthcare, etc. all people living the high life with no worries? work being done! US! celebrating?

The man thinking about co sm, monarchy, republic, etc. is the same humanity! man! infinite! being! unique! as so his possibilities aligned with the infinity? peace! where judgement conceals it?

有些人可能会更倾向于使用、采纳并依赖DeepSeek人工智能，认同它的判断，符合他们的偏好。

如果东大将世界观强加给他人，那就有问题了。这侵犯了他人的自由！这是暴力！是评判！是控制！

别误会我！难道你们不觉得伊甸园或天堂就类似一种GC主义体系吗？用诸如“福利”“社会保障”“全民医疗”这些词来描述。所有人都无忧无虑地过着高品质生活，各项工作都能完成。美国！在庆祝什么呢？

思考GC主义、君主制、共和制等的，都是同一类人！人类！无限的存在！独一无二！就如同他的可能性与无限相连。

Rob Roy

When something is a great deal or too good to be true that is when you need to look at it the closest, to see if anything else unknown is hidden in the deal. They may have created the greatest thing since sliced bread but based on China's past performances and who they hang out with and call their friends (like putin), China could be giving away bars of gold and I still wouldn't touch it. In case they had imbeded something else inside it.

What we have learned in these last years is that China always seems to have a covert agenda to all it does, and they don't do anything to promote a correction in their ideology.

China knows it could become a peacemaker and help bring peace and end Russia ‘s bullshit war, yet they choose not to do so. Their new tech is then just more of its hidden agendas, that need to be watched, not embraced no matter how good it looks.

当某样东西好得超乎想象，让人觉得不太真实时，你就得格外仔细审视，看看这背后是否隐藏着不为人知的猫腻。他们（中国）或许创造出了自切片面包以来最伟大的东西，但鉴于中国过去的表现，以及他们结交并称其为朋友的那些人（比如普京），就算中国免费送金条，我也不会碰。就怕里面藏着别的东西。

这些年我们了解到，中国似乎做任何事都有隐秘的意图，而且他们也不采取任何行动来修正自身的意识形态。

中国明明知道自己可以充当调解人，助力实现和平，结束俄罗斯那场荒谬的战争，可他们却选择不作为。所以，他们的新技术不过是其更多隐秘意图的体现，无论看起来多好，都应该保持警惕，而不是欣然接受。

Freddie Chen

Because that AI is not a tool for biased people like you who think if the answer did not match his/her expectations, it is the biased propganda.

Like Zionism think if you did not support it, you were anti Semitic.

AI are hel people in a lot of industries that will not differ Chinese or US but giving an answer applies to objective fact. For exmple, 1+1 equals to 2.

It is just a waste of you to use AI but not for other people.

因为该人工智能不是为像你这样有偏见的人服务的。你这类人觉得只要答案不符合自己的预期，就认定是有偏见的宣传。

就像犹太复国主义者认为，只要你不支持他们，你就是反犹主义者。

人工智能在许多行业帮助着人们，它不会区分是中国人还是美国人，只会给出符合客观事实的答案。例如，1 + 1 就等于 2。

对你来说，使用人工智能是浪费，但对其他人并非如此。

Stephen Ford

The questioner is misinformed.

DeepSeek is open source and can be trained using any data the users choose, so any bias they want, or no bias at all.

The significance of DeepSeek is the reported huge reduction in scale to achieve what other models have done. If true this represents a vast quantum leap in efficiency.

The lead DeepSeek currently has won’t last long because the techniques will be copied by everybody else.

To prattle on about bias is just sheer ignorance about DeepSeek’s significance, which is about AI technology, not how it is applied.

提问者的信息有误。

DeepSeek是开源的，用户可以使用自己选择的任何数据对其进行训练，所以既可以加入他们想要的任何偏见，也可以做到毫无偏见。

DeepSeek的重要意义在于，据说它大幅缩小了模型规模，却能实现其他模型所能达成的效果。如果这是真的，那意味着在效率上实现了巨大飞跃。

目前DeepSeek所具有的领先优势不会持续太久，因为其技术会被其他各方效仿。

一味喋喋不休地谈论偏见，完全是对DeepSeek重要意义的无知，它的意义在于人工智能技术本身，而非其应用方式。