8 Matching Annotations
  1. Last 7 days
    1. resemble their own positionality

      I agree. People tend to gather with their own "group" because it makes them feel safe. As a result, many are unwilling to spend the time and energy engaging with people from different cultural backgrounds, even though such interactions can broaden their perspectives and be highly beneficial.

    1. stochastic parrots

      The examples are solid. these groups barely show up in the data to begin with, and then the filters go and strip out what little they did post.

    2. with her gender (nurse, teacher)

      Gender-job stat's a solid example. nobody coded that bias in, the model just picked it up from the training text. Bringing in Stochastic Parrots is a nice touch since it ties this to a real scandal, not just theory. Also looking back to the Gramsci point, train on existing culture, you inherit its biases.

    3. large language models

      it's a fascinating analogy, taking a theory from nearly a century ago and applying it to LLMs feels genuinely fresh. The author invited us to look at what "cultural hegemony" might look like in the age of AI.

    4. because of cultural hegemony.

      I agree that culture plays an important role in keeping social systems in place. Schools, the media, and other institutions shape the way people think about society and what they see as "common sense." As a result, people may accept the existing system without even realizing how much these cultural influences affect their views.

    5. who weren’t exposed to much media or culture.

      I agree that Europe had a rich cultural tradition, but I think the author's description of Russia may be an oversimplification. Even though many people were illiterate and had limited access to mass media, they were still influenced by local culture, religion, and community networks. Culture is broader than formal education or the media.

    6. LLMs are inherently conservative technologies

      The author argues that LLMs mainly reproduce dominant cultural values because they are trained on existing texts. Using Gramsci's concept of the historic bloc, the author explains that these values become embedded in AI models and are difficult to change. However, I wonder if this is always the case, if LLMs will continue to evolve through updated data, and user interactions, if so, they seem to be shaped not only by the past but also by continuous social and technological change.

  2. Jul 2026
    1. digital cosmopolitanism

      Digital cosmopolitanism builds on Anderson's idea of imagined communities, where print media helped people imagine belonging to a nation. It extends this to the digital age, arguing that the internet lets people form communities across national borders based on shared interests and values.