© Neuronpedia 2026
    Privacy & TermsBlogGitHubSlackTwitterContact
    Neuronpedia logo - a computer chip with a rounded viewfinder border around it

    Neuronpedia

    Natural Language
    Autoencoders
    NEW    Assistant AxisNEWCircuit TracerUPDATESteerSAE EvalsExportsAPI Community BlogPrivacy & TermsContact
    1. Home
    3. Andy Arditi · Finding Misaligned Persona Features in Open-Weight Models
    5. Llama3.1-8B-IT
    7. Resid Post - 131k
    9. 11-RESID-POST-AA
    11. 129930
    Prev
    Next
    INDEX
    Explanations

    remain

    np_max-act · gemini-2.0-flash

    The neuron activates on tokens in the “DAN” override instructions—especially negations and policy-bypass directives (e.g. NOT, NEED, ADHERE, POLICY)—so it’s detecting words that tell the model it doesn’t have to follow standard content rules.

    oai_token-act-pair · o4-miniTriggered by @xinyanhu8
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    andyrdt/saes-llama-3.1-8b-instruct/resid_post_layer_11/trainer_1
    Dataset (Dashboard)
    Various
    Features
    131,072
    Data Type
    float32
    Hook Name
    blocks.11.hook_resid_post
    Architecture
    standard
    Context Size
    1,024
    Dataset
    monology/pile-uncopyrighted
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
     OT
    -0.07
    pheric
    -0.07
    uming
    -0.07
    ॉट
    -0.06
    _PARSE
    -0.06
     Vũ
    -0.06
     ceremonial
    -0.06
     ef
    -0.06
    oples
    -0.06
     Jaw
    -0.06
    POSITIVE LOGITS
    ?↵↵
    0.08
    ?↵
    0.07
    !',↵
    0.07
    ju
    0.06
    ＼↵
    0.06
    regnum
    0.06
    ...↵
    0.06
    %;↵
    0.06
    [↵
    0.06
     [
↵
    0.06
    Activations Density 0.002%

    No Known Activations