INDEX
    Explanations

    News articles/headlines

    This neuron reliably activates on the leading words of news‐style headlines, especially proper names or entities at the start of a title.

    New Auto-Interp
    Negative Logits
    viously
    -0.07
    .sdk
    -0.06
     liner
    -0.06
     Slav
    -0.06
     мг
    -0.06
     Balanced
    -0.06
    management
    -0.06
     Sim
    -0.06
    277
    -0.06
     sexually
    -0.06
    POSITIVE LOGITS
     بت
    0.07
    0.07
    ораль
    0.07
    خوان
    0.06
     ");
    0.06
    .Category
    0.06
    Palette
    0.06
    0.06
    ορειο
    0.06
     fetching
    0.06
    Act Density 0.062%

    No Known Activations