Neuronpedia logo - a computer chip with a rounded viewfinder border around it

    Neuronpedia

    INDEX
    Explanations

    one

    np_max-act · gemini-2.0-flash

    The neuron detects the appearance of the “condone / promote” refusal phrasing (i.e. the split tokens cond + one and promote) in the assistant’s disclaimers.

    oai_token-act-pair · o4-miniTriggered by @xinyanhu8
    New Auto-Interp
    Negative Logits
    Escape
    -0.07
    .Cascade
    -0.06
     search
    -0.06
     setObject
    -0.06
    -community
    -0.06
     looping
    -0.06
     Stay
    -0.06
     "','"
    -0.06
     guitars
    -0.06
    -----------
    -0.06
    POSITIVE LOGITS
     způsob
    0.07
    gg
    0.07
    هه
    0.07
    HEL
    0.07
    UTH
    0.06
    дан
    0.06
    ops
    0.06
     fov
    0.06
    대회
    0.06
     marching
    0.06
    Act Density 0.010%

    No Known Activations