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

    Neuronpedia

    INDEX
    Explanations

    Katie

    np_max-act · gemini-2.0-flash

    personal names

    np_max-act · gpt-5-miniTriggered by @yooniel31

    tokens that are code identifiers or placeholder names (capitalized/underscored identifiers like NAME_*, function/class names and other programming symbols).

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

    Explanation could not be parsed.

    eleuther_acts_top20 · gpt-5-miniTriggered by @yooniel31

    The neuron activates on numeric tokens and variable‐style placeholders (e.g. digits like “1”, “7–9”, and markers like “NAME_”), i.e. it’s looking for numbers or parameter placeholders in the text.

    oai_token-act-pair · o4-miniTriggered by @yooniel31
    New Auto-Interp
    Negative Logits
    UNKNOWN
    -0.07
    .Div
    -0.07
     dư�
    -0.07
    .gwt
    -0.07
     CEL
    -0.07
     انتظ
    -0.06
     �
    -0.06
     �
    -0.06
     setState
    -0.06
     OSS
    -0.06
    POSITIVE LOGITS
    iale
    0.08
    čan
    0.07
    (&___
    0.07
     bahwa
    0.07
     něj
    0.07
    ј
    0.07
    čů
    0.06
    čka
    0.06
    ــ
    0.06
     στη
    0.06
    Act Density 17.598%

    No Known Activations