© 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
    2. Gemma-3-27B-IT
    3. 31-GEMMASCOPE-2-TRANSCODER-262K
    4. 31594
    Prev
    Next
    INDEX
    Explanations

    modelmodelThe user wants a concise explanation (3-20 words) of a neuron's behavior, based on the provided lists.The explanation should identify a pattern or concept without using specific forbidden phrases.Let's look at the lists:* **MAX_ACTIVATING_TOKENS**: `date`, `details`, `details`, `details`, `details`, `specifics`, `details`, `details`, `methods`, `sizes` * This list contains words related to attributes, properties, or characteristics of something. Multiple mentions of "details" strongly point to this.* **TOKENS_AFTER_MAX_ACTIVATING_TOKEN**: `isn`, `about`, `about`, `are`, `of`, `of`, `are`, `are`, `vary`, `aren` * This shows what typically follows the activating tokens: "details isn about", "details about", "details are", "specifics of", "methods are", "sizes vary". This reinforces the idea of describing attributes.* **TOP_POSITIVE_LOGITS**: `திரை`, `México`, `בח`, `Further`, `などを`, `operated`, `జాగ్ర`, `反复`, `画布`, `財` * This list contains a mix of languages and symbols, with a few English words. It doesn't immediately reveal a clear semantic pattern that aligns easily with the other lists. It might indicate the neuron triggers for non-English text or certain obscure English words, but the other lists are much stronger indicators.* **TOP_ACTIVATING_TEXTS**: * "after my training period. (My training data has a cutoff, but the exact date isn't publicly specified)." - mentions `date`.This neuron detects information about model attributes

    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_31_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
    😼
    0.39
    tble
    0.39
     ஆனால்
    0.38
    phia
    0.36
    intestin
    0.35
    山上
    0.35
     ⠀
    0.35
     Clk
    0.35
    gram
    0.35
     tj
    0.35
    POSITIVE LOGITS
     திரை
    0.38
    México
    0.37
     בח
    0.36
     Further
    0.36
    などを
    0.35
     operated
    0.34
     జాగ్ర
    0.34
    反复
    0.34
    画布
    0.34
    財
    0.34
    Activations Density 0.013%

    No Known Activations