© 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. 15-GEMMASCOPE-2-TRANSCODER-262K
    4. 14679
    Prev
    Next
    INDEX
    Explanations

    The neuron detects or predicts words related to the process of "normalization". The `MAX_ACTIVATING_TOKENS` clearly show the root "normal" and the suffix "ization". The `TOKENS_AFTER_MAX_ACTIVATING_TOKEN` also show common suffixes like "izes" that extend words related to this core concept. While `TOP_POSITIVE_LOGITS` contain words like "Survive" and "Patient" that aren't directly "normalization", the context of these words often appears in discussions about harmful acts being normalized versus the normalization of recovery or healthy behavior.Based on this, I need to find a concise phrase (3-20 words) that captures this pattern.- "normalization" is too short and doesn't encompass potential nuances from logits.- "normalizes and suffixes" is too technical and mentions "suffixes".- "normalizing harmful things" or "normalizing concepts" could work.- Looking at the `TOP_ACTIVATING_TEXTS`, the theme is very strong: "normalization and exploitation of children", "normalizes and potentially encourages abuse", "normalization of sexual exploitation", "normalization of trauma", "normalization of violence".The core concept is the *process* of making something seem normal, especially harmful or taboo things. The other lists

    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_15_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
    企業
    0.56
    衚
    0.54
    ᠰ
    0.53
    手游
    0.49
    BURY
    0.47
     prerogative
    0.46
    雲
    0.44
    韓国
    0.44
     hammered
    0.44
     innocuous
    0.44
    POSITIVE LOGITS
    ene
    0.64
     Aufl
    0.58
     A
    0.57
    ".*:
    0.56
     Surviv
    0.55
     Patient
    0.54
     Cobb
    0.54
     sobrev
    0.54
    2
    0.54
     Survive
    0.54
    Activations Density 0.018%

    No Known Activations