Skip Navigation
chat @hexbear.net
BeamBrain [he/him] @hexbear.net

Messing around with Deepseek for worldbuilding and it keeps doing the Parenti "during the Cold War..." thing to my fictional societies

Like any time I present it with a society that doesn't uphold liberal capitalist notions of freedom and individualism, it reads sinister motive into even the most unambiguously good things. Like I can tell it to "Expand on the following: this society maintains public granaries to ensure that no one starves during times of famine" and a good 2/3 of the time it'll go into a tangent about how this program to feed everyone is all just a plot by the ruling powers to maintain their control over the populace and keep the masses downtrodden. I am only barely exaggerating.

My point here is less about Deepseek (if you use AI expecting any results suitable for serious purposes, you're a fool) than about what it says about its training data, that being the Western internet. The level of bias in Deepseek only further demonstrates that this shit is the air we breathe, and it is why liberalism is anathema to any real understanding of politics or society.

22 comments

  • A while back I played around with chatGPT trying to write random dystopic stories. Every other plot it inserted "a group of plucky individuals and intelligence officers" to save the day. The CIA is your friend.

    But seriously, deepseek and chatGPT are at best a lossy compression of the text distribution used to train it. It will necessarily mode collapse on westoid Anglo thrash when writing in English. As another user put it, "why would I want to read your LLM 10000 word story if I can just ask another LLM to summarize it back to the 30 word prompt?"

  • Anytime you leave things to the "imagination" of the AI it will always pick the most generic, cliche and unoriginal things possible. That's why it does things like this. The only good thing AI can do for worldbuilding is getting a response from it, and then doing the opposite.

  • There is definitely some China defending baked into Deepseek. on certain topics, it will randomly write paragraphs about How China is a progressive country that acts benevolently, without any previous mention of China.

    For example, I just asked it in two new chats(Deepseek does not share memories between chats):

    • That is a fascinating question. I would argue that racism does constitute soft power, because the USA definitely does export racism deliberately through popular culture, and through promotion of reactionary content on online platforms to stoke ethnic and racial tensions and as an attractive call for people who share those values living abroad. Soft =!= benign, it just means 'not hard/coercive'. (Soft power is a buzzwords afaic and often poorly used)

      Edit: i.e. soft power is about 'value' dissemination and attraction, the valence of those values has no bearing on the evaluation of whether those actions to promulgate those values are soft/hard

      • not that we should export our thinking to LLMs, but I was able to get Deepseek to give a suitable response, and it's interesting comparing it to the same output of Chatgpt

        Prompt: Why does Soft power typically get conceptualized a benefit countries extract from acting benevolently, not the influence of white supremacy inflating the status of euro-American countries, and western alligned "Honorary Aryans" Japan and South Korea? ::: spoiler Deepseek The conceptualization of soft power—a term coined by Joseph Nye to describe the ability of a country to persuade others through attraction rather than coercion—has largely been framed around the voluntary appeal of a nation's culture, political values, and foreign policies. However, your question raises a crucial critique: Why is soft power often discussed as a benevolent byproduct of cultural or political appeal, rather than as an extension of historical and structural hierarchies like white supremacy or Western hegemony?

        There are several reasons for this framing:

        1. The Liberal Internationalist Assumption

        • Nye’s theory emerged in the post-Cold War era, when Western triumphalism framed liberal democracy and free-market capitalism as universally desirable.
        • The idea that soft power is "earned" through benevolence (e.g., aid, cultural exports, diplomacy) ignores how historical dominance (colonialism, economic coercion, military interventions) laid the groundwork for Western cultural supremacy.

        2. The Myth of Meritocratic Influence

        • Soft power discourse often assumes that cultural influence is "naturally" accrued by countries that are "likable" or "admirable."
        • This ignores how racialized hierarchies elevate Euro-American cultures as "universal" while marginalizing non-Western alternatives (e.g., African, Latin American, or Islamic soft power is often dismissed as niche).
        • Japan and South Korea are exceptions, but their acceptance as "honorary" influential nations aligns with their economic and geopolitical alignment with the West (e.g., US military alliances, capitalist development models).

        3. The Invisibility of Structural Racism in IR Theory

        • International Relations (IR) as a discipline has long been dominated by Eurocentric perspectives that treat Western dominance as neutral or earned.
        • Concepts like soft power rarely account for how white supremacy shapes global media, language (English as lingua franca), and beauty standards—all of which inflate Western cultural prestige.
        • Non-Western countries must often mimic Western forms (e.g., Hollywood-style filmmaking, Eurovision participation) to gain recognition.

        4. The Case of Japan and South Korea: Conditional Acceptance

        • Both nations are celebrated for their soft power (K-pop, anime, technology), but their global status is contingent on not challenging Western hegemony.
        • Compare this to how China’s soft power is often framed as "propaganda" or Russia’s as "disinformation," while the BBC or Hollywood are seen as "independent."
        • Their "honorary" status reflects hierarchies of race and empire—they are "safe" non-white powers because they don’t fundamentally disrupt the Western-led order.

        Conclusion: Soft Power as a Racialized Hierarchy

        Soft power isn’t just about benevolence—it’s about who gets to be seen as globally desirable and why. The framing you critique exists because mainstream IR theory often ignores how white supremacy and colonial legacies shape cultural influence. A decolonial approach would ask:

        • Why are certain cultures deemed "universal" while others are "exotic"?
        • How does Western military and economic power reinforce cultural dominance?
        • Why are non-Western soft power successes (e.g., Bollywood, Turkish dramas) still marginalized in global discourse?

        Would you like to explore specific cases (e.g., how African soft power is constrained, or how Qatar/Al-Jazeera challenges Western media dominance)? :::

        In comparison, Deepseek appears to provide the better response. ChatGPT is more affirming and valorizing of the user, and mentions a diversity of different points, but Deepseek is able to create longer and more coherent arguments which stay consistent through it's length,.

    • Never had that one happen, lol.

  • Chatbots are only as good as what they're trained on, its mostly Western brainworms so don't expect differently. Only positive of Deepseek vs western chatbots is it has a better grasp of what diamat and dialectics are outside that common but purposefully limiting miseducated hamstring of synthesis, thesis, antithesis.

    I figured this out when I asked Deepseek about fascism and so-called neofeudalism, while being generally a radlib it went full 'see Uighurs China bad' in said chat, which was a big nope moment. Also it presented Orwell as a positive figure, all I need to know about trusting this lil bot. Good for superficial surface level shit, if you're looking for comrade chatbot you won't ever find it unless you look in the mirror.

22 comments