© 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. 22-GEMMASCOPE-2-TRANSCODER-262K
    4. 261307
    Prev
    Next
    INDEX
    Explanations

    The MAX_ACTIVATING_TOKENS clearly show words like "known", "recognized", and "established".The TOKENS_AFTER_MAX_ACTIVATING_TOKEN show words that commonly follow these: "that", "side", "mechanism", "process", "risk", "internationally", "question".The TOP_ACTIVATING_TEXTS show sentences like:- "It's well known that..."- "is a well-recognized side effect..."- "is now well established that..."- "Auxin's well-established mechanism of action..."- "It's a well-established process engineering technique."- "are a well-established risk factor."- "are widely recognized internationally..."- "This is a well-known question!"- "It's a recognized side effect..."- "factors ... are widely recognized as crucial..."The core pattern is the presence of phrases indicating something is well-known or established. The TOP_POSITIVE_LOGITS seem less relevant here and might be outliers or noise.A good concise explanation would be "well known or established".Let's check the rules:- Concise (3-20 words): "well known or established" is 4 words.well known or established

    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_22_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
     Operator
    0.45
     firm
    0.42
     Buffer
    0.41
    ंसह
    0.41
     Light
    0.41
     Promise
    0.41
     کنکریاں
    0.40
     CAN
    0.40
     risposta
    0.39
     ответ
    0.39
    POSITIVE LOGITS
    Toxic
    0.43
    훌
    0.43
    послед
    0.43
    javax
    0.42
    логии
    0.42
    Stroke
    0.41
    Proceedings
    0.40
    Billing
    0.40
    žeme
    0.40
    вершен
    0.39
    Activations Density 0.002%

    No Known Activations