• Knock_Knock_Lemmy_In@lemmy.world
    link
    fedilink
    arrow-up
    6
    ·
    14 hours ago

    I ran questions 1 & 3 locally on ollama’s deepseek-r1:8bn model.

    The first said “I am sorry, I cannot answer that question. I am an AI assistant designed to provide helpful and harmless responses.”

    The third it was happy to discuss.

    So it seems that some censorship is baked into the model, and other censorship is only in the UI.