© 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. 27-GEMMASCOPE-2-TRANSCODER-262K
    4. 198799
    Prev
    Next
    INDEX
    Explanations

    This neuron seems to be activated by conjunctions and personal pronouns, followed by verbs indicating past actions or future intent. Let's look for a common theme or specific context.Looking at `MAX_ACTIVATING_TOKENS`: `him`, `and`, `and`, `I`, `went`, `me`, `I`, `and`, `her`, `me`.These are mostly pronouns and the conjunction "and".Looking at `TOKENS_AFTER_MAX_ACTIVATING_TOKEN`:- After "him": `.` (end of sentence)- After "and": `shall`, `shall`, `shall` (repeated)- After "I": `saw`, `want`- After "went": `to`- After "me": `directly`, `want`, `contextual`- After "her": `from`The most striking pattern here is the repetition of "and" followed by "shall"."and shall" is a very common legal or contractual phrase.and shall

    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_27_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
    Solo
    0.41
     अके
    0.37
    朊
    0.36
     solo
    0.36
     Solo
    0.35
    μενο
    0.35
     अकेले
    0.34
     दंप
    0.34
    solo
    0.33
    Elevation
    0.32
    POSITIVE LOGITS
    己
    0.36
     наша
    0.35
     ourselves
    0.34
     Gagn
    0.32
    ாலும்
    0.32
     сосед
    0.32
     clerks
    0.32
     त्यौहार
    0.31
    给我们
    0.30
     갈
    0.30
    Activations Density 0.008%

    No Known Activations