© 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-3-27B
    5. 31-GEMMASCOPE-2-RES-262K
    7. 105995
    Prev
    Next
    INDEX
    Explanations

    soap, taking selfies, what enrages

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

    medical conditions (vomiting, meconium, stool, constipation)2. corrections (grammar, mistakes)3. discussions and personal topics4. bodily functions (composition of farts)The `TOP_POSITIVE_LOGITS` are in foreign languages, which suggests the neuron might be picking up on abstract properties of words rather than just English semantics, or it's multilingual.The most consistent pattern seems to be related to bodily functions and related contexts. bodily functions and conditions

    np_acts-logits-general · gemini-2.5-flash-lite
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    google/gemma-scope-2-27b-pt/resid_post/layer_31_width_262k_l0_medium
    Prompts (Dashboard)
    392,802 prompts, 256 tokens each
    Dataset (Dashboard)
    monology/pile-uncopyrighted
    No Configuration Found
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
    0.54
     (
    0.50
    knock
    0.48
    middle
    0.48
    (
    0.47
     K
    0.46
     H
    0.46
     gain
    0.45
     regional
    0.45
     middle
    0.44
    POSITIVE LOGITS
     bếp
    0.54
     содержи
    0.54
     кухни
    0.54
     спраши
    0.52
    Celebrating
    0.52
     दिखाते
    0.51
     аласыз
    0.51
     αφού
    0.51
    𒀯
    0.49
    式の
    0.49
    Activations Density 0.001%

    No Known Activations