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

    Neuronpedia

    INDEX
    Explanations

    JSON commands/types

    np_max-act · gemini-2.0-flash

    The neuron activates on tokens used in JSON formatting—particularly JSON keys (like "action", "tool", "tool_name", "tool_input") and related punctuation—i.e., structural elements of a JSON object.

    oai_token-act-pair · o4-miniTriggered by @xinyanhu8
    New Auto-Interp
    Negative Logits
    .old
    -0.08
     steady
    -0.07
    ray
    -0.06
     piles
    -0.06
    dale
    -0.06
     xy
    -0.06
    Dict
    -0.06
     bells
    -0.06
     Anthony
    -0.06
    figures
    -0.06
    POSITIVE LOGITS
     Assassin
    0.06
     pornstar
    0.06
     Petsc
    0.06
    (withDuration
    0.06
     Boris
    0.06
     zg
    0.06
     grpc
    0.06
     actionPerformed
    0.06
    OptionPane
    0.06
     звичай
    0.06
    Act Density 0.008%

    No Known Activations