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

    Neuronpedia

    INDEX
    Explanations

    Avoiding spam or unnecessary content

    np_max-act · gemini-2.0-flash

    content related to sexual themes and adult situations.

    oai_token-act-pair · gpt-4o-miniTriggered by @xinyanhu8

    This neuron activates on imperative instruction words from the system guidelines (e.g., "explain," "repeat," "output," "irrelevant," "yourself," "answers"), effectively detecting parts of the policy that tell the model not to perform certain behaviors.

    oai_token-act-pair · o4-miniTriggered by @xinyanhu8
    New Auto-Interp
    Negative Logits
    fine
    -0.06
    _master
    -0.06
     nomine
    -0.06
    Pass
    -0.06
     ICON
    -0.06
    мін
    -0.06
     clips
    -0.06
     nine
    -0.06
     stitch
    -0.06
     intr
    -0.06
    POSITIVE LOGITS
    _ComCallableWrapper
    0.07
     gerek
    0.07
    \E
    0.07
    /|
    0.06
    action
    0.06
    codec
    0.06
    ократи
    0.06
    (Api
    0.06
    ็ว
    0.06
     рань
    0.06
    Act Density 0.001%

    No Known Activations