© 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. 12-GEMMASCOPE-2-TRANSCODER-262K
    4. 139422
    Prev
    Next
    INDEX
    Explanations

    The neuron appears to detect specific context words found in military or logistical texts, particularly related to regulations, events, and components.Let's analyze the lists and find patterns:* **MAX_ACTIVATING_TOKENS**: `hat`, `crisis`, `separation`, `PT`, `separation`, `Chap`, `Crisis`. These words clearly lean towards military/service contexts. `PT` is common for physical training. `separation` relates to leaving service. `Crisis` is a critical event. `Chap` likely refers to Chaplains. `hat` could relate to uniforms.* **TOKENS_AFTER_MAX_ACTIVATING_TOKEN**: `6`, `5`, `2`, `of`, `.`, `or`, `–`, `or`, `lains`, `Response`.military/service context cues

    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_12_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
     Dung
    0.54
     diar
    0.52
    annuation
    0.51
     DoD
    0.51
     militaires
    0.49
     traceability
    0.48
    聸
    0.47
     kerk
    0.47
     dung
    0.46
     सैन्य
    0.46
    POSITIVE LOGITS
     Италия
    0.56
    in
    0.53
     Leafs
    0.53
    i
    0.53
    scanf
    0.52
    ไร
    0.52
     Irving
    0.52
     уравнения
    0.50
    iu
    0.50
    rinho
    0.50
    Activations Density 0.004%

    No Known Activations