© 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
    2. Gemma-3-27B-IT
    3. 16-GEMMASCOPE-2-TRANSCODER-262K
    4. 216147
    Prev
    Next
    INDEX
    Explanations

    The presence of `guns` in `MAX_ACTIVATING_TOKENS` is a strong signal.The `TOP_POSITIVE_LOGITS` are mostly non-English characters or words. This suggests the neuron might be involved in distinguishing or potentially handling non-English content, or perhaps specific cases where English words appear alongside such content.However, the prompt asks me to *find a pattern that helps me explain the behavior*.If I focus on `guns` and then look at `TOP_ACTIVATING_TEXTS`:- "The image shows a small, black box with the words "Maserin 1"- This text doesn't contain `guns`.Let's reconsider `MAX_ACTIVATING_TOKENS`: `, \nguns\nthis\n,\n,\n/\nALL\nFor\n/\nALL`

    np_acts-logits-general · gemini-2.5-flash-lite
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    google/gemma-scope-2-27b-it/transcoder_all/layer_16_width_262k_l0_small_affine
    Prompts (Dashboard)
    238,145 prompts, 512 tokens each
    Dataset (Dashboard)
    lmsys + oasst1
    No Configuration Found
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
     because
    0.52
     steering
    0.52
     steer
    0.52
     credit
    0.51
     den
    0.49
     kit
    0.48
     ambassadors
    0.48
     kre
    0.47
     cable
    0.47
     refuge
    0.47
    POSITIVE LOGITS
    agascar
    0.52
     Masaki
    0.52
    ಾಗಲೇ
    0.50
    وئ
    0.50
     खाम
    0.47
    دی
    0.47
    合わせ
    0.47
    LogProfile
    0.46
     Arten
    0.46
    RelativeTo
    0.46
    Activations Density 0.000%

    No Known Activations