INDEX
    Explanations

    The neuron spots the document’s topical header token—typically a single uppercase label like “Football,” “Baseball,” “Coronavirus,” or “Soccer” immediately followed by a colon.

    New Auto-Interp
    Negative Logits
     _
    -0.08
     également
    -0.07
    thing
    -0.07
     tức
    -0.07
    /pop
    -0.07
     infantry
    -0.07
     kdo
    -0.06
    -0.06
    afort
    -0.06
     хто
    -0.06
    POSITIVE LOGITS
    рис
    0.07
    ilos
    0.06
    _dbg
    0.06
    atural
    0.06
    events
    0.06
     fille
    0.06
    elf
    0.06
    INY
    0.06
    loading
    0.06
    Flexible
    0.06
    Act Density 0.002%

    No Known Activations