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

    Neuronpedia

    INDEX
    Explanations

    AI self-description using "but" and "I"

    np_acts-logits-general · gemini-2.5-flash-lite

    sentences or passages where the assistant introduces itself or describes its identity, training, capabilities, and availability.

    oai_token-act-pair · gpt-5-miniTriggered by @yooniel31

    The neuron is essentially flagging the assistant’s own “long‐form” explanation turns (the multi‐paragraph, bullet‐list responses) as opposed to user utterances. In other words, it turns on for tokens in the model’s detailed breakdowns.

    oai_token-act-pair · o4-miniTriggered by @yooniel31
    New Auto-Interp
    Negative Logits
    我們要
    0.51
    !!!!!!!!!!!!!!!!
    0.49
    Fuck
    0.47
     vimos
    0.46
     нам
    0.46
     fucking
    0.45
    !!!!!!!!
    0.45
     consiglio
    0.45
     imo
    0.44
     veamos
    0.44
    POSITIVE LOGITS
     Developers
    0.63
     Capabilities
    0.63
    Unlike
    0.60
     OpenAI
    0.59
    Developers
    0.59
     myself
    0.58
     developers
    0.58
    Capabilities
    0.57
     capabilities
    0.56
     Unlike
    0.56
    Act Density 0.462%

    No Known Activations