© 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. 59-GEMMASCOPE-2-TRANSCODER-262K
    4. 58829
    Prev
    Next
    INDEX
    Explanations

    abbreviationsLet's re-evaluate if it could be Method 2 with a slight stretch.TOKENS_AFTER_MAX_ACTIVATING_TOKEN: f, PR, DR, DR, dR, OR, SR, PR, OR, SRDo they start with the same letter? No.Is there another pattern? They are all short, mostly 1 or 2 letters. Are they *all* single letters? No. Are they *all* 2-letter strings? No.Method 3 seems the strongest.TOP_POSITIVE_LOGITS are AR, CR, ER, BR, FR, GR, MR, TR, SR, IR. They are all 2-letter strings.This pattern (2-letter strings) is observable and explainable

    np_max-act-logits · gemini-2.5-flash-lite
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    google/gemma-scope-2-27b-it/transcoder_all/layer_59_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
    r
    0.47
    Gr
    0.47
     র
    0.47
     r
    0.45
     ر
    0.45
    Cr
    0.44
     Gr
    0.43
    Ber
    0.43
    Br
    0.43
    Wir
    0.43
    POSITIVE LOGITS
     AR
    1.67
     CR
    1.65
     ER
    1.60
     BR
    1.60
     FR
    1.58
     GR
    1.53
     MR
    1.51
     TR
    1.48
     SR
    1.46
     IR
    1.45
    Activations Density 0.005%

    No Known Activations