• niklas@lemmy.world
    link
    fedilink
    English
    arrow-up
    15
    ·
    6 hours ago

    They train grok like other llm Models, but they use system prompts to manipulate answers. That way they can tweak it quickly without retraining.