© 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. 10-GEMMASCOPE-2-TRANSCODER-262K
    4. 207541
    Prev
    Next
    INDEX
    Explanations

    **Explaining Neuron Behavior**My goal is to synthesize the provided lists into a concise phrase (3-20 words) that explains what the neuron detects or predicts by finding patterns.Based on the `MAX_ACTIVATING_TOKENS` (Greece, China, Rome, civilizations, Greek, history, Romans, centuries) and the `TOKENS_AFTER_MAX_ACTIVATING_TOKEN` (which are predominantly punctuation like '(', '-', ':', ','), a clear pattern emerges. The neuron seems to activate on terms related to historical civilizations. The tokens following suggest these terms are often used as headings or introductions to further information. The `TOP_POSITIVE_LOGITS` contain words from various languages, which can sometimes indicate a capacity for broad categorization or cross-lingual association, though here the pattern in `MAX_ACTIVATING_TOKENS` is much stronger. The `TOP_ACTIVATING_TEXTS` confirm this, showing phrases like "Ancient Greece (6th Century BCE...", "Origins & Early Discoveries (China - 9th - 13th Centuries)", "Ancient Greece: The term...", and "ancient Rome" followed by descriptions.The core theme is historical civilizations.Therefore, a fitting explanation would be:ancient civilizations and history

    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_10_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
     American
    0.61
     America
    0.59
     advantage
    0.59
     atrophy
    0.53
     vibes
    0.52
     AMERICAN
    0.52
     of
    0.51
     that
    0.51
     aime
    0.51
     american
    0.50
    POSITIVE LOGITS
     மாற்ற
    0.55
     путем
    0.54
    un
    0.52
    calculation
    0.52
    ᱔
    0.51
     Пусть
    0.50
     செய்து
    0.48
     تھے۔
    0.48
     —,
    0.48
     Держа
    0.48
    Activations Density 0.051%

    No Known Activations