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

    Neuronpedia

    Jacobian LensNEW
    Natural Language
    Autoencoders
    NEW    Assistant AxisNEWCircuit TracerUPDATESteerSAE EvalsExportsinterp-engineNEWAPI Community BlogPrivacy & TermsContact
    1. Home
    3. Gemma-2-27B
    5. 22-GEMMASCOPE-RES-131K
    7. 26527
    Prev
    Next
    INDEX
    Explanations

    asking what's wrong

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

    The neuron flags question phrases that ask “what’s wrong” or “why … wrong,” i.e. it activates on question words (What/Why) paired with “wrong” or similar error-seeking terms.

    oai_token-act-pair · o4-miniTriggered by @jyhe0408
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    google/gemma-scope-27b-pt-res/layer_22/width_131k
    Prompts (Dashboard)
    24,576 prompts, 128 tokens each
    Dataset (Dashboard)
    monology/pile-uncopyrighted
    No Configuration Found
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
    without
    -1.07
     without
    -1.03
    various
    -1.00
     χωρίς
    -0.99
     a
    -0.98
     diinginkan
    -0.91
     ohne
    -0.91
     tergantung
    -0.89
    льную
    -0.88
     вопросам
    -0.85
    POSITIVE LOGITS
     why
    1.85
     that
    1.34
     Why
    1.23
    能让
    1.22
     почему
    1.21
    why
    1.07
     настолько
    1.05
    为什么
    1.02
    Why
    0.99
     khiến
    0.99
    Activations Density 0.046%

    No Known Activations