© 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 EvalsExportsAPI Community BlogPrivacy & TermsContact
    1. Home
    2. Gemma-3-27B-IT
    3. 18-GEMMASCOPE-2-TRANSCODER-262K
    4. 89109
    Prev
    Next
    INDEX
    Explanations

    The neuron's behavior is strongly associated with geographical identifiers related to historically significant regions, particularly those in the Middle East. The presence of "Galilee," "Palestinian," "Palestine," and multiple instances of "Land" in `MAX_ACTIVATING_TOKENS`, combined with textual references describing the "Holy Land," "Lake of Galilee," and "first-century Palestine" in `TOP_ACTIVATING_TEXTS`, points to this neuron detecting concepts or contexts involving this specific geography.The `TOKENS_AFTER_MAX_ACTIVATING_TOKEN` list shows common continuations like punctuation or descriptive words ("was", "from", "Desert"), which are typical after place names or nouns. The `TOP_POSITIVE_LOGITS` list seems less directly relevant to this specific pattern, showing a mix of foreign language fragments and technical terms, which might indicate this neuron isn't *exclusively* about the geographical aspect but has broader connections or is being probed with diverse data.However, the primary and most consistent pattern is clearly related to the geographical region.Therefore, the explanation should capture this core association.**Explanation:** holy land regions

    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_18_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
     natal
    0.44
    nay
    0.43
     தான
    0.42
    曲
    0.42
     yep
    0.42
     रेन
    0.42
     guna
    0.41
     arp
    0.41
     curved
    0.41
     জীববিজ্ঞান
    0.41
    POSITIVE LOGITS
     නිෂ්පා
    0.52
    нії
    0.48
    ﮭ
    0.46
    щиеся
    0.46
     Etats
    0.45
    ranno
    0.44
    Preprocessing
    0.44
    国内外
    0.43
    itesi
    0.43
    Approval
    0.43
    Activations Density 0.000%

    No Known Activations